Alternative political views get downvoted (Puh-lease!) because they are unpopular views: An analysis of downvote collapses related to Charlie Hebdo

May 17, 2015 Laura Hale charlie hedbo, quora

Disclaimer: This was not as anonymized on the belief that a lot of Quora is built on reputation from answering questions. If you see a name that you think should be anonymized or want your name removed, suggest the edit and it will be removed. All of the data used to build this was publicly available on Quora.

Apologies: There are one or two charts and graphs here. If you use a screenreader, you can’t really “read” these. I apologize for that. Any solution would be appreciated.

Ooops: Yeah. I have a huge spelling error in my graphs. 😦 At the time of publishing, I could not get Quora to allow me to upload more things, format headers and process urls.

Introduction

Sometimes uncomfortable topics need to be explored. First, it seemed like people said that “Downvote Cabal exists everywhere on Quora!” I feel like I have killed this boogeyman. The next boogeyman to appear recently has been, “Unpopular political views are frequently collapsed! Quora should do something about these downvoters punishing views they do not want!”

I’ve asked several people for data who have made this assertion, because I am a rather trusting soul and I could totally believe that particularly narrative. Alas, none of them provided any support for their claims. Thus, in a fit of pique, I have decided to address this.

I’ve chosen the Charlie Hebdo topic because I’ve seen enough questions on Quora that appear to be “The terrorists were justified right?” and “We should put limits on freedom of speech to not insult religion right?” and “What the terrorists did was wrong, but didn’t Charlie Hebdo bring it on themselves?” and… well, I’m rather angry about those sentiments. I’m actually more than a little annoyed, but that’s neither here nor there. The important thing on Quora is there are some excellent people doing an excellent job explaining from my POV as to why these questions are faulty. I don’t need to spend huge amounts of time and energy with there being something wrong on the Internet.

So now I am left to ask: Are these Quorans who think that unpopular political views result in downvote collapses? How often is this occur on a hugely important social event currently taking place? Is what is occurring insideCharlie Hebdo (newspaper) different than other topics? Do these people who are complaining about this have a leg to stand on?

Before beginning this analysis, it should be made implicitly clear that the content of the non-collapsed answers were not examined to see if they represented a diversity of views on Quora. It is possible that Quora has a certain group think on some subjects, and that manifests itself by people holding other political views not answering political questions in general, or only answering where existing questions’ answers indicate their answer would fall on like ears.

Methodology

This methodology draws largely from The Great Quora Downvote Cabal does not exist: Characteristics of collapsed answers.

http://www.quora.com/Charlie-Heb… was visited and all the questions were included. A few additional questions were used after searching for Charlie Hebdo on Quora’s search. The total number of questions included was 141. A full list of the questions can be found in the Appendix.

In looking for collapsed answers, two separate spreadsheets were created. The first contained the following columns: Number, Question, Number of answers, Number of collapsed answers, Total collapsed.[1] The reasons from The Great Quora Downvote Cabal does not exist: Characteristics of collapsed answerswere used, and a new column added as required for new reasons. Additional columns for each collapse reason to get a more refined understanding of causes of collapses.

Number here is just the order that the question was viewed in and to make it easy to cross reference later. It has no significant meaning. Total answer is the number Quora displays. It includes only the number of non-collapsed answers.

The second spreadsheet specifically looked at the actual answers and had the following column headings: Question, Author, Number of followers, Gender, Upvotes, Views, Word count, Answer question?, View expressed, Quote. The first is the number assigned to the question itself with number corresponding to the question on the first spreadsheet. Author is who wrote the question. Number of followers is how many followers the author has. Gender is if they are male, female or anonymous. [2] Upvotes is how many upvotes the collapsed answer has. Views is the number or views the question has. Word count is answer length and was determined using Word Count Tool – Free Tool to Count Number of Words. Word counter! . The answer was copy pasted to derive it. Date written is when the answer was written. Answer question is did the answer actually answer the question as asked. This was based on interpretation of the author, with every goal of being neutral. The options were Yes, Somewhat and No. Somewhat was for the category of answers that sort of addressed the question but not completely . View expressed was based on an interpretation of the answer to determine where the answer fell politically. The categories were created as each response was written. Quote is the text of the answer. This is a large number of variables.

This data was gathered on January 14. Because of the high interest in this issue, the data is very fluid and subject to change. Answers that are not collapsed at the time of this analysis may still be collapsed. New answers may still be written, and additional questions may be asked.[4]

Results

106 of the 141 questions, or 24.8% of all questions, had zero downvotes. UsingThe Great Quora Downvote Cabal does not exist: Characteristics of collapsed answers as a baseline, this number was 11.3% higher than the Quora norm of around 13.5%.

42 of the 64 collapsed answers were a result of downvotes. This 65.6% is in line with the baseline comparison, which was slightly higher at 66.0%. 5.2% of all answers on this topic were downvote collapsed, with 803 total answers of which 739 were non-collapsed. This is less than the baseline of 5.6%. With 7.9% of all answers collapsed, it is again less than the baseline of 8.5%.

What were the reasons for the collapse? Downvotes accounted for 42. Joke answer accounted for 9 collapses. Answer comment, and Does not answer question accounted for 7 each. Unverified name and Violates image policy accounted for 2 each. Lacks details, Very poor formatting that significantly distracts from readability, Needs to be in English and BNBR violation account for 1 each. The numbers do not match the baseline, but there is a large difference in numbers of collapsed answers, 256 verses 64. This can account for some of the discrepancy.

If the two data sets from Hebdo and the by topic analysis are combined for reasons for collapse, you get the following graphs.

Collapsed answers as a function of all answers are relatively small, accounting for 8% of all answers.

When downvote collapses are separated from other collapse reasons, they account for a relatively significant percentage of collapses: 5% o 211 of all sampled Quora answers were downvote collapsed, while the remaining answers, all 109, account for only 3% of all total sampled Quora answers. These other reasons are admin collapsed based on contributor submitted reports. The combined breakdown can be best seen in the graph below, with the 8 joke answers accounting for 3.75% of all collapsed answers.

Amongst downvote collapses for the Hebdo population, there were 25 male answers, 6 female answers and 11 anonymous answers. With women making up 19% of answerers, this outside the norm for established gender representation on Quora of between 27% and 39%. [3]

Women were more likely to have their answers downvoted on this subject than in general. Anonymous users were also more likely to be downvote collapsed. Men were less likely to be downvote collapse by a significant percentage.

If anonymous users are ignored, 27 authors account for 31 answers with 4 authors having two answers collapsed. These authors include Syed Anwar, Craig Hubley, Derek Larkin and Sandra Li.

There is no evidence of an active downvote cabal targeting users for this topic based on the number of downvotes. This is supported by observational data for an anonymous user with 4 of the exact same copy paste answers that were downvoted, which appeared on other answers but were not collapsed.

Downvote collapse answer writers on average had 87 followers, a median of 24 followers and a mode of 1 follower. The user with the most followers and a downvote collapse was Nitin Gupta, with Usman Qazi having the second highest at 397. Authors with large numbers of followers were not getting downvote collapsed. A user with 1,000 followers was higher than the baseline sample. That overall had a downvote collapse answer writers with an average of 43 followers, a median of 10 and a mode of 1.

The average number of upvotes per collapsed answer was 0.75, with a median of 0 and a mode of 0. The average was higher than the baseline, but the median and mode were the same.

The average number of views per collapsed answer was 312, with a media of 49 and mode of 25. While the average was close to the baseline of 312, the median and mode were significantly as the baselines were a median of 108 and mode of 129. Collapses answers generally are not getting many views.

The long anonymous copy paste killed the average length of answers. With them included, the average answer 218 words, with a median of 60 and a mode of 1,069. If these copy-paste answers are removed, the numbers change dramatically to an average of 128, median of 57 and mode of 3. In all cases though, this is different than the baseline which had an average answer wordcount of 46, median of 18 and mode of 1. The Charlie Hebdo answers were longer than other downvote collapses, but is still shorter than the average answer on Quora, which Writing longer content for upvotes established at 324 words and median of 128 words. Shorter answers still lead to collapses.

The relationship between views and answer length are noticeably different. The ratio of views to word count is not proportional. The correlation for Charlie Hebdo downvote collapse answers is -0.17. It is almost completely random. This is distinct from the downvote collapses from the baseline.

In general, answer length played less of a role in downvote collapses for Hebdo answers as there was less of a correlation, gender was less of a factor in that gender was more evenly split, having few followers was less of a factor as people had the more followers on average than the baseline, and upvotes was not a factor because it was very close to the baseline. The two groups of downvote collapse answers were similar, but distinct. Answers were also less likely to be downvote collapsed for Hebdo answers than the baseline.

With the background knowledge of how Hebdo downvote collapses and how that contrasts with the baseline, the following results can be better understood.

Did the collapsed answers address the question? This was rather evenly split, with 14 yes, 15 no and 12 somewhat. This split invites almost as many questions as it answers. On the whole though, it suggests those that did not answer the question as asked fared worse than those who did.

For the answers that answered the question, they were an average of 64 words and a median of 43. Compared to the whole of the Charlie Hebdo answers, this suggests that shorter answers that answer the question were punished.

Though then again maybe not. When the copy-paste answers are left in, the average answer length for answers that do not answer the question is
312 words but when we take away those four answers, it drops down to 60. This is below the average length for yes answers. The same pattern holds true for the mean. This could suggest that short answers that do not answer the question are punished more.

The answers that somewhat answer the question have an average length of 271 words and a median of 129 words. This appears to confirm a thought that longer answers that do not directly address the question are bad, and suggests including those copy-and-paste downvote collapses are legitimately part of the no group.

Longer answers that do not answer the question result in an increased likelihood in this topic of getting a downvote collapse.

Now for the all important, why this research was done question: The political opinions expressed in collapsed answers. Based on the assumptions of a number of Quora community questions, answers and comments, Quora has a hive mind. The expected result is a uniformity in political opinion of collapsed answers.

Neutral accounts for the largest number. These were largely answers that asked a non-political question and got a direct answer back, namely translating French phrases.

The next largest category was anti-Islamic statements. These were answers that basically suggested all Muslims were terrorists, should be kept out of Europe, etc.

If “conspiracy”, “anti-American” and “anti-American, conspiracy” were added together, they would actually total 7 and be equal to anti-Islam views. These political views are not exactly the same, or are at least not espoused by the same people. They would most likely crossover in the anti-semitic group so there is potentially that.

On the whole, this could confirm that fascist, racist and xenophobic, anti-free speech political views tinged with anti-western democratic ideals, conspiracy theories and anti-Americanisms are more likely to be subject downvotes than other political views. It is just a rather broad group that is hard to lump into one unified political group. If one assumes that these can all be broadly lumped together for ideological purposes, then it is feels safe to assume that yes, some ideological views are more likely to be downvoted than others. On the other hand, if you assume that they cannot be lumped together and that the moderates, neutrals and self promotional bloggers are important, then that argument is much weaker. The second interpretation is much more compelling.

What is another approach to checking this? What do the answers that address question as asked have as their politics when they were downvote collapsed?

Politics Yes African POV 1 Anti-American 1 Anti-establishment, Anti-rich 1 Anti-intolerance 1 Anti-Semitic 1 Anti-victim 1 Anti-West 1 Neutral 5 Pro-victim 1

From here, it is much harder to argue that politics alone are the major factor resulting in downvote collapse on political questions. Neutral answers are the single biggest group. If grouped with pro-victim and African POV, then these outnumber other political type answers that would be viewed as problematic. Some of these neutral collapses that answered the question as asked also had comparatively high views.

On Quora, there is a bias against anonymous answers and questions. Their downvote collapse answers were equal to women in the general selection, and accounted for a quarter of all downvote collapses here. On 8 of the 28 questions with collapsed downvote answers, the only answers that were downvoted collapses only contained anonymous answers. Their political views were not hugely diverse but did fall into the problematic range that could suggest collapse for those reasons.

Political views Anonymous Anti-Semitic 2 Anti-American, Conspiracy 3 Anti-Islam 4 Conspiracy 1 No true Scotsman 1

The only problem with that narrative is only one of those answers actually answered the question as asked. That political view is one of the two anti-Semitic answers.

Conclusions

The conclusions for The Great Quora Downvote Cabal does not exist: Characteristics of collapsed answers said, ” one of the easiest ways to avoid getting your answers collapsed is to avoid writing short answers, especially when they add little value. Also, get followers. People with more followers appear to have fewer collapsed for downvoting answers. ” It also said, “Nothing in this analysis though shows any signs of a number of unexplained outliers and multiple people showing up with downvotes.” It goes on to say no successful downvote cabal appears to exist.

These same conclusions appear to hold true for this situation, and this particular political topic. The answers are similar enough in terms of length, upvote totals, follower numbers for answer writers that if they were put into the same dataset as the original, you’d get very little variance. The major difference actually appears to be the gender of the people getting downvoted and the number of anonymous who were downvoted.

Answers that on first blush could have the implication that they were downvoted solely because of the politics of their views just do not stand up to scrutiny when other variables are examined.

Your politics alone are not getting your answers downvoted. I am growing increasingly convinced after a second analysis that while there may be fine differences between certain topics in the percentage of gender collapses, and percentage of collapses based on topic that overall, these collapses are not based on ideology, point of view or downvote cabals. Rather, these collapses are based on people finding an answer, downvoting it (often instead of reporting) because it is short, does not address the question, written by someone who has few followers so less social pressure not to downvote collapse, and was written by a male. To a small degree, downvote collapses likely represent an actual quality control mechanism.

More research may be needed on the topic. If you disagree with the conclusions here, you are invited to use https://wikinewsreporter.files.w… . This is a list of both data tables for all general, and Hebdo downvote collapse tables. See if you draw a different conclusion, or use similar methodology for selecting questions for inclusion and add more collapses to it to see if they support these conclusions.

Footnotes

Participant observer. During this I made a rule generally to not upvote or downvote content. There was an uncollapsed answer by anonymous. I downvoted it because it was a copy and paste from an article from a conspiracy theory site. It had appeared on various answers before and did not address the question. Sometimes it was collapsed. Sometimes it was not. The report collapsed it on downvoting. Ooops. It was the only time where in collecting data I reported or downvoted an answer. I did the new collapse. After that, I made sure that I followed the rule of not reporting.
The terms male/female, man/woman are used interchangeably. Apologies to anyone offended by this. In a Quora context, this refers to gender and which people chose to present themselves as on Quora.
For the research I did at ’15 Top Writers: Topic Area Expertise: Who knows and what do they know? between December 31 and January 5, there are more male than female writers, with 67.7% (103 total) being male and 32.2% (49 total) being female. This involves a sample of 152 top writers out of 220 identified ’15 top writers.
Somethings that do not occur to you until after you get into your data collection: In this particular case, it might have also been instructional to look at the comments on downvote collapses. Unlike other topics I have observed, this one appeared to have more comments on downvote collapses then I have seen elsewhere.

Written 14 Jan.

The ’15 Top Writer upvote benefit: It is not as big as you think

May 17, 2015 Laura Hale quora, top writer

Apologies: There are one or two charts and graphs here. If you use a screenreader, you can’t really “read” these. I apologize for that. Any solution would be appreciated.

Introduction

I feel cynical. I often feel like when I browse Quora Community is hear complaints in answers or in questions that are basically, “Top Writers are so ebil! They get 1,000 upvotes on average, even on their crap throw away one liners! What’s wrong with Quora???” My sympathies, they exist not.

I want a thousand upvotes on average. Then I could use all my credits… to… er… well, based on some questions about women on Quora, I think use my feminine wiles to trade in my credits for a new car, new laptop, money to travel to cover disability sports and lots of credit for Candy Crush Saga from some unsuspecting male Quora user. Wait. No. Quora credits do not give that.

What is the value of Quora upvotes and credits anyway beyond a form of social recognition that you said something that people felt the instant gratification compulsion to spend half a second upvoting?

I digress. There is this myth in certain parts of Quora that popularity through the Top Writer program translates into answers with huge numbers of upvotes. Like lots and lots of upvotes. Also, thousands of upvotes on all your questions. But how true is this statement? Do 10,000 followers translate into an average of 1,500 upvotes per answer? This little analysis seeks to find out how many upvotes an average ’15 Top Writer answer gets, and what the conversion rates for followers to upvotes is.

Methodology

A ’15 Top Writer profile page was visited. The total number of followers they had was recorded on a spreadsheet. Next, from the author profile page , the answers page on the sidebar was selected. From there, the page was scrolled down to try to make visible all the answers the ’15 top writer had answered in the past month, generally defined here as between December 10 to January 11. Because of how Quora displays its information, this is not an exact science, and trying to make sure only answers from exactly inside that period were collected becomes less precise because of time constraints for writers with more than 75 answers in that period. Once the page displayed these answers, the page contents was copied and pasted to notepad to rid it of its formatting, and then copied from there to a blank excel spreadsheet. The column was then sorted to put the upvote count list in about the same place. All other rows that did not contain upvote totals were eliminated. The total number of answers in the period was then recorded. This was then copy and pasted back to notepad, where ” upvotes by ” and ” upvote by ” were replaced with tabs. There was also the need to search for and convert answers with thousands of upvotes from the 1k format to the 1,000 format. The raw number was then pasted to a unique column for each writer, which already contained total answers and followers. The extra column was then eliminated. This took between 2 and 8 minutes for each writer on the list. Because of this, the data size is limited.

The decision to use only answers from the past month was based on the idea that it would be primarily collecting recent answers, without worrying about the historical impact of older answers. It was also an issue of time. Once the total answer total hits about 400, it takes about 6 to 8 minutes to process because of the need to scroll down a huge amount and because it takes extra time to sort out rows that do not belong as sort puts “1 upvote” before “1 zebra”.

A major problem with this methodology is in the upvote count when it comes to answers with 0 upvotes. As a result, answers is the 0 upvotes are not included, both for math purposes and in the question total. This is because Quora does not include “0 upvotes” as a line when copying and pasting. This means that the 0 upvote answers are not counted. Based on a sample involving one user, the number most likely to be impacted is the mean, with median and mode unlikely to be effected.[5] It is unlikely to have a huge impact on the overall narrative, with the results likely to be that there is a slight increase in upvote totals.[6]

The selection of ’15 Top Writers included was not entirely random from the list. the first writers that were selected for analysis were ones that appeared on my feed. Others were selected from Who are the Top Writers with the fewest followers? and Which Quora Top Writer had the fewest followers and answers upon being named? and similar questions in order to try to improve the pool of ’15 top writers who have few followers. How many Top Writers 2015 are teenagers? was also visited to expand the diversity a bit. After this, a few authors were chosen at randomish from Top Writers 2015: Who are the 2015 Top Writers? . The goal was to include 28 or more writers as this represents about 10% of the known ’15 Top Writer class, and to have a diverse selection from within this subset.

This data was collected on January 10 and January 11.

Results

Of the 39 class of ’15 top writers sampled, the average number of upvotes amongst all answers in the last month[1] was 32, with a median of 8 and a mode of 1.

Of the 39 sampled authors, they had an average of 8,840 followers and a median of 4,078. The writer with the fewest followers was Yedu Krishnanwith 238, and the writer with the most was Jimmy Wales with 98,300.

As a group, they were semi-active, with average of 91 and a median of 38 answers during this one month period. The most prolific author was Jon Mixon with over 500.[1] The least prolific was Clayton C. Anderson with 5. Marco North, with 8, was the only other sampled writer with 10 or fewer answers in the past month.

When the median upvotes per answer based on median upvotes per writer is looked at, the median is 6[4]. The lowest median number of upvotes per answer is 1, a number set by Dan Birchall. The second lowest is Laura Hale,Yedu Krishnan, Jason MacDonald and Brady Postma with 2. At 2.5 upvotes on the median per answer is Susie Downing and Sunil Kumar Gopal. With 3 upvotes as the median on a per answer basis is Jesse Lashley, Vivek Mehta andBen Fraser . Birchall has 2,103 followers. Amongst the people with 2, the follower counts are 424, 2,012, 238, and 1,724. For 2.5, the follower counts are 1,993 and 6,233. For 3, the follower counts are 5,943, 3,887 and 1,422. These are not small numbers of followers, so large follower counts do not guarantee a lot of upvotes.

The highest on the median is 419, and is set by Jimmy Wales. Clayton C. Anderson is second at 160. Robert Frost is third with 97. See? Unless you’re Jimmy Wales or working for NASA (and even that second one is a close but not sure guarantee), you do not get 100 upvotes per answer as a rule.

A number of users have a mode number of upvotes at 1. These users includeLaura Hale, Anne W Zahra, Josh Manson, Jesse Lashley, Dan Birchall, Yedu Krishnan, Susie Downing, Vivek Mehta, and Sunil Kumar Gopal. The highest mode was Jimmy Wales with 1,100 upvotes. Robert Frost had a mode of 23, while Claire J. Vannette was third with 19. Finishing up amongst other high performers were a pair with a mode of 14, including Barry Hampe, and Xu Beixi. Again, with the exception of Wales, this number is way below 100. The consistent high performance with these authors is not there.

The average Ratio Followers:Upvotes was 426 while the median was 761. In other words, most followers are not blindly upvoting content of top writers. The best mean ratio was set by Yedu Krishnan at 1.7 upvotes for every follower.[2] Clayton C. Anderson had the best ratio on the median, with 51 followers for every upvote. The worst performer on the average for converting followers to upvotes was Sunil Kumar Gopal who had an average of 2037 followers for every upvote. He was also the worst on the median with 2493 followers for every upvote.[3]

Much of the discussion around unfair advantage for Top Writers (Quora program) is that the large number of followers leads to an unfair competitive advantage in getting upvotes. The graph below has the average, median and mode number of upvotes per author during the period of December 10 to January 11 against the writer’s follower count.

Only two data point exists above the 1,000 mark for upvotes. An additional 6 are above the 100 mark for upvotes, two of which are median and four are means.

What is the correlation between average number, median number and mode number of upvotes? 0.89, 0.90, and 0.86 respectively. If Jimmy Wales with almost 100,000 followers and Robert Frost with almost 55,000 followers are removed as outliers, these correlations change dramatically. The correlations become 0.10, 0.35, and 0.67 respectively. In essence, outside of the super power users with more than 50,000 users, the relationship between the total number of upvotes and the number of followers you have is random, mostly random and probably not entirely random but since the maximum mode number is 19, it isn’t a particularly high number.

Conclusion

Being a ’15 Top Writer and having a lot of followers does not guarantee you hundreds or thousands of upvotes unless your name is Jimmy Wales. If you think that being a Top Writer or that having thousands of followers is a ticket to hundreds or thousands of upvotes for every answer you write, this is just not conceivably true. Nothing in the data at all supports any sort of conclusion of this sort. Normally I would say that this subject could use further research to explore the topic in more depth but in this case, I am not sure what the added value in doing much more research on this would do. I would encourage others to do so, but only because I think more research on this subject would be useful in debunking this myth.

Footnotes

Footnotes are not in the order that they appear in the text. Rather, they are in the order they were written.

Jon Mixon is hugely prolific, and when done being hugely prolific, he is hugely prolific. His data only includes the period between December 23 and January 10. This accounts for 444 answers. The amount of scrolling down required to go further a lot more time, and the ROI for getting his average number of upvotes was limited. I do not believe the lack of his full data skewed the outcomes for this analysis.
With 41 answers and 278 followers, Yedu Krishnan ‘s outliers of 3,800, 693, 863, especially since he had 19 answers with only 1 upvote.
With 6,233 followers, Sunil Kumar Gopal average 3 upvotes per answer, had a media of 2.5 upvotes per answer, and a mode of 1.
The number is derived by calculating the median number of upvotes for each user. After a user median is calculated, the median for that new data column was then found. This is different then getting the median for the whole array of upvotes. If the average of the median scored is used, you get 24.6. A preference is for median here over mean because the outliers here appear to really distort perspective because quantity does matter as the premise of many questions about unfairness for top writers is that alltheir answers get this benefit.
In the case of my own answers at http://www.quora.com/Laura-Hale/…, I had 34 answers with 0 upvotes in the described period. This downcounts my activitiy from 191 answers to 157. My average upvote total goes from 3.6 to 2.9 when these 31 are count. The median and the mode stay the same.
I hope. Insecurity! Manual data mining error found when processing results.

Appendix

Full data set with all included writers and math.

Written 11 Jan.

The Great Quora Downvote Cabal does not exist: Characteristics of collapsed answers

May 17, 2015May 17, 2015 Laura Hale gender gap, quora

Apologies: There are one or two charts and graphs here. If you use a screenreader, you can’t really “read” these. I apologize for that. Any solution would be appreciated.

Introduction

If you take a quick look through Quora Downvote Cabal, you come away with the impression that there is an active community of users seeking to actively “punish” certain users and responses by downvoting their answers into oblivion. There are accusations of this behavior somewhat regularly. According to Marc Bodnick’s answer to Is Quora censored by strategic down voting user groups?, people sometimes complain to Quora about this alleged behavior going on and they have investigated a few of these claims. They have not found evidence to suggest that this is actually going on.

So what’s going on? Is there a successful Quora Downvote Cabal who can get answers collapsed? Who is getting targeted? What content is getting targeted?

The phrase successful and “get answers collapsed” is important in this question because that is what matters. Quora uses an algorithm, which they apparently continually fine tune, that determines relative placement of an answer. This placement is highly unlikely to be primarily determined by the number of downvotes, which would account for answers being placed lower. Stephanie Vardavas’ answer to Why does Quora allow cost free, no explanation, anonymous down voting? explores this issue somewhat, and I’ve taken a stab at answering it at Laura Hale’s answer to Why does Quora allow cost free, no explanation, anonymous down voting?.

This piece is interested only in answers that have been collapsed, and more specifically collapsed for downvoting.

Transparency and Moderation: March 2012 Changes by Marc Bodnick on The Quora Blog addresses some of the issues related to collapsing in an important way. Bodnick says, “Historically, Quora Moderation has uncollapsed certain answers. In the interest of transparency, we are publishing a new policy on uncollapsing downvoted answers. In summary: “Quora Moderation may uncollapse a downvoted answer if it appears likely that the collapse is related to theanswer representing a minority point of view. Answers that are uncollapsed should be well-written and thoughtful.”” Bodnick’s answer is reinforced with What is Quora’s policy on uncollapsing downvoted answers?.

Any Down Vote Cabal is going to need to work to get around that potential issue if they are downvoting with the intention of suppressing opinions.

Given the evidence, it appears to be a largely baseless assumption that a downvote cabal could be defined as successful unless they are able to collapse an answer through downvoting. This action also needs to be understood as completely different to reporting answers and users, which can also lead to answers being collapsed.

Methodology

To find questions, https://www.quora.com/sitemap was used as the starting point based on topic. Questions were then sought with 5 or more answers. This minimum threshold was chosen under the belief that it would increase the statistical chances of finding a collapsed answer.

In looking for collapsed answers, two separate spreadsheets were created. The first contained the following columns: Topic, Number, Question, Number of answers, Number of collapsed answers, Total collapsed. Additional columns for each collapse reason to get a more refined understanding of causes of collapses.

Number here is just the order that the question was viewed in and to make it easy to cross reference later. It has no significant meaning. The topic tag is based on what Quora topic the question was found under using the sitemap. Sometimes, it appears wildly out of sync with what the question actually asks. Total answer is the number Quora displays.

In selecting questions based on topics, https://www.quora.com/sitemap was visited. A topic was selected. To provide a somewhat random selection of questions included, the questions were chosen by going to the questions list for a topic. For Startups for example, https://www.quora.com/Startups/a… was the list. This list was then scrolled down until a question appeared that had 5 or more answers. The question was then included. Without doing the research and using only this list, it appears that only 1 to 10% of questions actually qualified as they had the requisite number of answers. Some topics had more unanswered questions or having questions with few answers than others. In using this process, new questions or questions with recent answers were the ones included. Thus, the results more likely reflect recent behaviors.

If a question was placed in 2 or more categories, it was counted in the first one it appeared in. In at least two cases, the tagging was so completely off that they were modified to remove inappropriate tagging.

A full list of the questions can be found in the Appendix.

The second spreadsheet specifically looked at the actual answers and had the following column headings: Topic, Question, Author, Number of followers, Gender, Upvotes, Views, Word count, Date written. The first two are from the question itself, with number corresponding to the question on the first spreadsheet. Author is who wrote the question. Number of followers is how many followers the author has. Gender is if they are male, female or anonymous. Upvotes is how many upvotes the collapsed answer has. Views is the number or views the question has. Word count is answer length and was determined using Word Count Tool – Free Tool to Count Number of Words. Word counter! . The answer was copy pasted to derive it. Date written is when the answer was written. This is a large number of variables. As the data is being mined by hand, it means a small sample that may not be entirely representative. Additionally, some things like scrolling to the bottom of questions takes a huge amount of time.

The data was gathered between January 8 and January 9. This data is very fluid and subject to change. Answers that are not collapsed at the time of this analysis may still be collapsed. What are the most profitable ways of making money online? was looked at on the night of January 8 for the first table. When it was gone back to on January 9 for the second table, another answer had been collapsed. In another case, when viewing a question page, the question actually said some one was currently writing an answer.

The decision was made to include 200 answers as this appeared to be a reasonable sample size, and would give a better idea as to the collapse answer to question ratio than trying to aim for a number of collapsed answers, as the second was something to which I had much less control over.

Results

In 200 questions sampled, 37 had collapsed answers, with 27 of those questions having answers collapsed because of downvotes. 13.5% of all questions had downvote collapses. Generally speaking, this number appears to be rather low and does not suggest a widespread culture of downvoting on Quora.

The total number of answers amongst the 200 questions sampled was 2,988. The total number of collapsed answers was 256, with the total number of downvote collapsed answers being 169. 8.5% of all answers are collapsed, and 5.6% of all answers are collapsed because of downvotes. 66.0% of all collapsed answers are because of downvotes.

In terms of other reasons for collapse, “Does not answer the question” was the second largest category with 25, “Unverified name” was third with 22 collapsed answers, “Needs to attribute sources/Better attribution/Needs attribution at the top” was fourth with 9 answers, Violates image policy was fifth with 7 collapses, “Does not disclose relevant affiliations” was sixth with 6 collapses, “Answer comment” at seventh with 5 collapses, “Need improvement (predate reasons, or no further details)” at eighth with 4 collapses, “Joke answer” at ninth with 3 collapses, ” very poor formatting that significantly distracts from readability” at ninth with 2 collapses, and “Lacks details”, “Needs to be in English”, “The use of a meme in an answer not specifically asking for one is no longer allowed in new Quora policy.” and “False information” each having 1 collapse.

In the introduction, there was the question of who was getting collapsed for downvotes. The second database includes 165 of the 169 downvoted answers. The missing four are from questions with large number of downvotes. The number of tabs open resulted in suboptimal computer performance, resulting in occasional accidental close of a window in these circumstances.[1] The 165 collapsed answers were made by 151 unique authors. 9.2% of these unique authors came from one user, “Anonymous”. Only one author, Randy Zeitman, had more than one answer downvoted and that total was 2.

Most people having their answers collapsed because of downvotes were have very few followers, have answers that otherwise are not getting upvotes generally have few views and are very short as evidenced in the table above. 9.1% of all collapsed answers were one word long.

50% of the answers were 18 words or shorter. To give this some perspective, the paragraph immediately above this has a word count of 44 according to the counting tool. We’re talking very, very short answers. This is an important and defining characteristic of collapses for downvotes: Short answers and one word answers.

To give word length additional perspective, Writing longer content for upvotesfound that the bottom quartile of Quora answers had median answers that were 21 words long.[4] Here, that 18 words is shorter but accounts for 50% of all the answers. The mode for the shortest 25% of answers was 9, compared to 1 here. The length of collapsed answers is well outside the norm for Quora.

The graph below gives a bit of perspective on a topic that is not being dwelled on much here. It is the relationship between views and length in words of collapsed answers.

In this case, the coefficient is -0.02948193. Basically, how many views your collapsed answer gets is a total crapshoot. Length is not a determining factor in views. Other variables not related are length of answer and number of followers. The correlation is 0.0990129. The relationship between upvotes on collapsed answers and total followers is also pretty random, with a correlation of 0.10659148. The relationship between length and upvotes of collapsed answers is also pretty random: 0.0446583. Collapsed answers are unique, especially when contrasted against no-collapsed answer length issues explored in the other analysis referenced.

People who have collapsed answers also appear to be new users to the site, as evidenced by having few followers. 35.7% of all answers were either written by anonymous users or by users with 0 or 1 follower.

What’s another characteristic of downvote collapses? They tend to be written by men. 137 were written by men, 14 by Anonymous and 14 by women. 83% of all downvote collapses were by men. If we subtract out Anonymous and only compare male:female,[2] that says 90.7% of all downvote collapses were written by men. This is well outside the norm for established gender representation on Quora of between 27% and 39%. [3]

It is actually difficult to account for this, without being willing to make large assumption based on content. These include that people who violate BNBR and get collapsed for this reason are more likely to be male, and that men are much more likely to use Quora to promote their products. The latter though may not always be the case as there were a a few of advertising type downvote collapses with names and pictures of pretty women. Their product promotion tended to be about a paragraph long, as opposed to male advertising accounts which often appeared to be 5 words long, followed by the link.

What topics have answers that get downvoted on Quora? The problem with this dataset is it did not intentionally wade intentionally into potentially cantankerous topics. Instead, it tried to provide a certain degree of randomness. Thus, it cannot really answer What kinds of answers or comments (in terms of topics covered or positions taken) are most likely to be subjected to downvote abuse for ideological reasons on Quora? as these questions largely did not show up during the sampling method used here.

The 200 questions in this sample came from 22 different topics, with 6 to 12 answers for each topic. At this point, it should be worth clarifying: Total answers and total collapsed answers are actually two different things. When Quora counts total answers that it displays on the page, it does not count downvotes.

Topics most prone to downvote collapses are design, mostly on the strength of a single question: What are the fastest growing websites?. This question may actually belong in another topic, but Quora topics are not always curated well. The next most popular topic for collapse is Startups at 9.7%, Science at 8.5%,International Travel at 6.8%, and Food at 5.5%. Seven have 0%, though realistically, you could put all the International Travel questions into Travel, and Startups into Business.

This does support a premise that some topics are more prone to downvotes than others. If you are trying to avoid a collapse, you may want to reserve your short, pithy one liners to Business, Entrepreneurship, Lean Startups, Photography, Travel, Writing, and YouTube .

How much of a factor is time? Without doing much of an analysis, some topics appear to have a long tail for getting answers while others had a short tail. In browsing, it became clear some topics were there and gone. This appeared to be particularly true for Dating and Relationships where looking for answers with downvotes seemed particularly challenging becausehttps://www.quora.com/Dating-and… appeared to mostly include questions that were posted within the past day if they had 5 or more answers. Despite that, when do downvote collapses happen?

The above graph appears to suggest that downvote collapses occur shortly after an answer is posted, within a day or two. Almost all the spikes visible are for the same questions, where multiple answers were written in the same condensed timeframe. There are two explanations. First, if there is not a cluster of people writing answers at the same time, it appears less likely that your answer will actually be collapsed. In this context, old questions are may be your friend to avoid getting your answer collapsed. Conversely, if your answer from a year ago has not been collapsed by now, it is unlikely to get collapsed in the future.

Contextual analysis
No formal contextual analysis was done as part of this analysis, though it is not hard to pick up some of the issues when looking at collapsed answers.

Why a few of these answers were downvoted seems obvious and a contextual analysis could really be beneficial if some one was doing further research.

Quotes from downvoted answers that probably violated BNBR:

Get a girlfriend soon.
Stop whining and grow up.
Consider that you may not be ambitious, talented and intelligent.
Jump off a bridge. You are wasting my oxygen.
yes,i admire to Hitler.
Fuck you
Which leads anyone with an IQ above room temperature to come to only one conclusion…
In a similar way that MSNBC unapologetically represents the view of the left-wing nut jobs.
Women are petty and vicious.
Lemme guess, you’ve read history only in your school books.
WRitiNg LiKe thIs?
In response to a question about the best science invention, “Disclaimer: This content is not Anti-Muslim; this is Anti-humanity !”
In response to a question asking if India inventing the plane: And the plane was hijacked by an Arab Guy to reach Jerusalem and Heavens.
And his followers made a berberic invasion on Bharat Bhoomi and destroyed the Airodrome.
now smoking is altogether harmless.go slip a cigarette and blow all your worries with the smoky puff.

Some answers that were downvoted also had a number of links, that said “See this site!” This was particularly common on the question for how to make money online. A classic example of this, “Click on link given below”. Marketing pitches on answers appeared to be a frequent candidate for this. A speculation content wise for this is that answers that are reported as spam are deleted, but there is no method of reporting an answer to say, “Reads like an advertisement for a product”.

Another characteristic of some downvoted answers was funky punctuation, poor grammar, types and spelling errors. An example of a complete sentence demonstrating this from a downvoted answer, “probablu there are statistics that entrepreneurual people leave sooner
and work less as programmers” Another example from an account using a business name, “Wat;s APP?” A third example, “As far as i feel it simply helps in bowel movement .also itworks as amorning supplement” Another example, “iwatch from apple company…….and the concept about the new iphone model i phone nano……..”

Some answers were not in English. Rather than report them, people appear to have downvoted them. One example: “Je suis Charlie ; vous êtes un bigot , avec une calotte et la barbe .”

Some answers that were downvoted appeared to be asking for clarification for the question instead of putting this as a comment in response to the answers. These downvoted and collapsed answers didn’t appear to offer any unique insight into why the question was problematic in the way it was phrased that added understanding to it. For example, “Please also tell how much honey to add and how much lemon to add.”

The problem with some of the easy contextual explanations for downvotes is there are too many examples of answers that easily fit into these contextual molds that do not get collapsed, and are written by people with similar numbers and answer views. Five word answers exists in some topics. The provide unsatisfying answers. Yet a five word answer for this question may not result in a collapse, where a similar answer has been collapsed on the same question or in a different topic.

Conclusion

At the end of the day, one of the easiest ways to avoid getting your answers collapsed is to avoid writing short answers, especially when they add little value. Also, get followers. People with more followers appear to have fewer collapsed for downvoting answers. While you may not get views for writing answers to questions clustered around other writers doing the same, doing so may also serve as a way to avoid getting your answer collapsed.

Nothing in this analysis though shows any signs of a number of unexplained outliers and multiple people showing up with downvotes. If there is a Great Quora Downvote Cabal, it is largely ineffective and not very broad in its focus. There would need to be a lot more substantial proof to support the existence of a large organized group of users regularly downvoting quality Quora content.

And obviously, more research is needed on this subject, larger data sets, etc. It would also be interesting to explore some of the characteristic Quora downvote collapsed answers in content and see how they differ from non-collapsed downvoted Quora answers. Such a contextual analysis coupled with a quantitative analysis would probably do a much better job at explaining what is actually going on.

Footnotes

Until Quora has an API and I have a programmer to datamine data I want, this will always be a risk. I do not believe this missing data negatively impacts the result as the clicking closures were randomly accidental, and as part of questions with large numbers of collapses.
The terms male/female, man/woman are used interchangeably. Apologies to anyone offended by this. In a Quora context, this refers to gender and which people chose to present themselves as on Quora.
For the research I did at ’15 Top Writers: Topic Area Expertise: Who knows and what do they know? between December 31 and January 5, there are more male than female writers, with 67.7% (103 total) being male and 32.2% (49 total) being female. This involves a sample of 152 top writers out of 220 identified ’15 top writers.
Writing longer content for upvotes was done on November 21. There were 36 women who had responses recorded and 94 men. That sample suggests women make up 27% of the Quora population.

Women and feminism: Male versus female answers was written on October 2. It found males answers made up 61.1% (258) of all answers, while females made up 38.8% (164).
For men. This doesn’t seem unfair as a comparison because there are so many men in this sample. If it is compared to women, the number looks worse.

Appendix

Written 10 Jan.

Upvotes: Your answer is not going to get the vote total you are looking for

May 17, 2015 Laura Hale quora

Disclaimer: Quick, dirty research done after finding a question in my draft space that I wanted to answer related to this. I find I prefer writing more extended blog entries on this subject than as separate answers because then I can keep my Quora research for fun centrally located and accessible for answering questions related to these topics at a later.

Introduction

On Quora, lots of users think that lots of people get huge numbers of upvotes. I’ve seen this attitude expressed in multiple places. It is often accompanied by a follow up question of, “Why am I not getting the same number of upvotes?” Some people are just hugely invested in the number of upvotes.

This is also a frequently occurring question on Quora. It has many variants that seek to explore this issue: What are the median and average number of upvotes per Quora answer?, What is the average number of upvotes for an answer to be good?, What is the average number of upvotes out of all answers on Quora?,What Top Writer has the lowest average number of upvotes per answer?, What is the average number of upvotes out of all answers on Quora?, Who has the most upvotes per answer on Quora?, Despite the growing number of users on Quora, why has the average amount of upvotes stayed the same?, Nobody upvotes my answers on Quora. Does that mean I’m stupid?.

There is something wrong on Quora, and it is the fact that people are not getting upvotes and this creates insecurity.

So what is the average number of upvotes on Quora? What should your realistic expectations be when you post an answer?

Methodology

On January 7, different questions were chosen at some what random but not really. It started with my own answer list to get questions. Two other sets of answers by specific users were looked at. Some random keyword searching for questions was also done. An effort was made to include India and USA centric questions as these user populations represent a large chunk of Quora readership. A complete list of these questions can be found in the appendix below.

A spreadsheet was created. A column was created for user gender, user name, total number of upvotes for a specific answer, and the question they answered. The goal was to get to around 500 answers. This number was chosen because it was a relatively large number, but still feasible to attain in the space of a few hours of manual data mining.

Gender was determined based on using one or more of the following criteria:

The person had a name traditionally associated with a specific gender from where their profile indicated they were from. Micheal is a traditional man’s name in the United States. Laura is a traditional woman’s name in the United States, England and Spanish speaking countries.
The person had a picture of themselves in their profile, and they appeared to be male or female.
The person said in their profile that they were male or female, or used male or female specific pronouns that indicated their gender.
The person answered a question where they indicated their gender.

Gender was chosen as the only other variable other than user because it is relatively simple to get. Some of the other research I have done suggests that there are a variety of variables at play for the total number of upvotes you can get. Writing longer content for upvotes looks at this with a much smaller dataset with content length as a defining variable. Women and feminism: Male versus female answers looks at this from a gender of answerer, and two topic perspective. Does Quora favor certain content over others? is another small dataset that looks at the issue based on topic.

The decision to focus only on these two variables was for three reasons: First, I am highly interested in how gender differences are manifested in Quora user behavior. Quora’s gender gap is personally interesting to me. Second, as stated previously, by limiting the variables, the data set can be larger. Third, most of the questions related to upvotes do not appear to be specific to these other variables.

Results

The 40 questions included 504 answers. Of these, 131 were written by women, 346 were written by men, and 27 were written by anonymous. If anonymous are not counted, this puts the percentage of female answers at 27.4%, and in the low range of other research I have done which finds the range of female participation in ranges between 27% and 39%. Broadly speaking, it may be reasonable to assume that in terms of gender participation, this Quora population is representative of the larger Quora population.

The average number of upvotes per answer was 15.4, with a median of 2 and a mode of 0.

The above graph gives an idea of the total number answers by upvotes. Most answers are clustered towards the left, having 16 or fewer upvotes. The answer with 1,000 and 1,500 upvotes are largely outliers. Over 120 of these answers have zero total upvotes: 24% of the answers sampled had zero upvotes. 19% of the answers had 1 upvote. 10% of the answers had 2 upvotes. 9% had 3 upvotes, and an additional 3% had 4 upvotes. 5% had 5 upvotes. Answers 5 or fewer upvotes accounted for 63.8% of all answers. Realistically speaking, that means an answer has a 2 in 3 chance of getting 5 or fewer upvotes, and that is not taking into account other potential variables such as the number of followers you have, topic popularity, your gender, and the length of your answer.

The following gives a slight better perspective on total upvotes. There is a huge range where there are voids for total numbers of upvotes. There is not a single answer with between 352 and 1,000.

The above graph also introduces the issue of gender. On Quora, there is sometimes a criticism by members of one gender or the other that by virtue of gender, one group is favored. Often this implication appears to include men saying that women have an inherent advantage on Quora because they get upvotes from men by virtue of being women that other men find attractive or upvote for the purposes of potentially dating. Conversely, there are women who have claimed that they are downvoted (and consequently not upvoted) more frequently than men because of some misogynistic elements on Quora.

The truth of it is when you do the math, it may be potentially advantageous to be female when answering. This advantage though is very slight and may account for a single upvote on an answer. If you’re looking for a variable you largely can control though in seeking upvotes, one thing you should avoid doing is publishing answers anonymously as this does hurt your ability to have highly upvoted outliers. If you’re looking for another reason to use your name, this is it.

Another variable that is easy to look at is the individual. These 504 answers were made by 424 different users. Excluding Anonymous, 6 different users had 3 or more answers picked up. These users were Laura Hale, Mystery Girl,Ânia Ferreira, Kent Fung, Anna Demers and John Colagioia. With the exception of myself with 22 answers, the total number of answers for each person is between 3 and 8. The answers included may not be 100% representative, but for now the assumption is going to be that they are.

Different authors have different patterns for upvotes, which does confirm the assertion that the median and average do not tell the whole story.

For a point of contrast, Mystery Girl has 60 followers. I have 416 followers.Ânia Ferreira has 3 followers. Kent Fung has 3,600 followers. Anna Demershas 3,000 followers. John Colagioia has 310 followers. Myself, Demers and Colagioia are current ’15 top writers. Fung was previously a top writer. Srivsastava and Ferreira are not top writers. Four are women and two are men. In this case, the male authors beat all the female authors on mean number of upvotes per answers. On median, only one woman beats the two men for median number of upvotes.

Half the writers beat the overall median for number of upvotes. All six of these writers fall below the median average for number of upvotes. Those 348, 1,000 and 1,500 outliers really do damage to everyone else’s average unless they are yours. In the case of these six, they clearly were not. For the mode, of the three people with them, two of them bettered the overall mode.

This simplistic analysis cannot explain why these individual patterns occur, but it does suggest that if you’re one of those authors who 2/3rds of your answers are getting 5 or fewer upvotes as the norm, there are other variables at play.

Conclusions

Most people do not get many upvotes. Also, many people do not get many upvotes. If people ask what the average number of upvotes is per answer, the best answer is: “Average is so deceptive. If you get three, consider yourself lucky as that is the Quora median. If a third of your answers get more than 5 upvotes, consider yourself very lucky. Now stop obsessing over this and go write quality answers.” If people ask why they should not be anonymous, explain to them that if they want credit through upvotes, they have a much better chance of getting a high performing answer if they use their real name.

Beyond those issues, the total number of upvotes an answer gets is a highly complex issue. Further research with larger datasets is needed to get a better grasp on the issue.

Appendix

* What do people in Western countries do to avoid spitting in public places?
* Are Australians the most travelled people in the world?
* For Americans living abroad: Has your view of the U.S.A. changed? How and why?
* Which universities in Australia require the SAT?
* What is the most dangerous situation you have ever been put in, while working on a story, as a journalist?
* If I want authentic foods when I go to a restaurant, is it safe to judge the quality and authenticity of the food by the number of people of that ethnicity who are dining there?
* What’s the biggest problem for a college student to learn English well?
* What is the best way to avoid English grammatical errors?
* What’s it like to be in a gay marriage?
* As of January 1, 2015, what percentage of Quora’s user base is female?
* What are the advantages and disadvantages of holding the Paralympics right after the Olympics?
* What should I keep in mind while planning to start learning Spanish as my first foreign language?
* What can journalists do if there is no news?
* What does it take to be followed by Top Writers?
* What are the perks of being your friend?
* How do I find out why I was born?
* How can I handle multiple interests/hobbies?
* How is time depicted in Bharatanatyam?
* I lost my father unexpectedly two weeks ago. Now all that I have in my life is my mom and my career. What is the road ahead for me?
* What are the commonly broken New Year’s resolutions?
* What is one thing you wish you could do with the points earned on Quora or in any game?
* Is it easy for me to learn German if I can speak French and a little bit of English already?
* How did you become confident and outspoken?
* Why are Korean women so good at golf?
* Why are there so many “how to learn xx language” questions on Quora?
* When is playing hard-to-get a bad choice?
* Which one responds best to the needs of tomorrow’s world, British Curriculum (BC) or International Baccalaureate (IB)?
* How can I be more patient when learning a language?
* Are there any useful apps to learn Portuguese?
* I told my boyfriend I’m not comfortable being sexually intimate at this stage of the relationship, and he told me I was being selfish, demanding and unfair to him because as a guy he has his needs. Am I really?
* What is the best kind of commitment you can make to a girl when you are at the age of 20?
* What are some differences between the different accents of Brazilian Portuguese?
* How could India become a superpower?
* Are cats evil?
* How much do Fannie Mae and Freddie Mac support the existence of the 30-year fixed mortgage?
* What should I look for in a mortgage?
* What effective strategies might there be for countering an “Ebola suicide bomber” threat?
* Are there any charities in Pakistan which give money to the families of suicide bombers?
* How do people outside the US view the contest between Romney and Obama?
* Is Mitt Romney a bully and a liar?

Written 8 Jan.

’15 Top Writers: Topic Area Expertise: Who knows and what do they know?

May 17, 2015 Laura Hale quora

Disclaimer: This was not as anonymized as well it probably could be on the belief that a lot of Quora is built on reputation from answering questions. If you see a name that you think should be anonymized or want your name removed, suggest the edit and it will be removed. All of the data used to build this was publicly available on Quora.

Also, again, network analysis is not my general thing but I feel more comfortable with it. If there is an error in my network analysis, please point it out.

Apologies: Male and female, man and woman are used interchangeably. I know there is a difference between gender and sex. In the context of Quora, we’re probably talking more about gender than sex, because the usage is based on how people present themselves online. No offense is intended.

Introduction

Quora has a lot of experts, a number of them recognized through the Top Writer program. They often write answers that provide new insights in areas that can be worth readings and rereading. This expertise, sometimes couched with humor, is often recognized through large followings.

The question is: How distributed is the knowledge base on Quora, expressed through this year’s selection of top writers? What areas is knowledge most concentrated? Is there a difference in male and female knowledge sharing by topic concentration, and in the breadth of individual expertise by number of topics people have answered questions in? How connected is this knowledge network to topic specialists in other areas?

Methodology

Drawing from a list of ’15 top writers[6], each top writer profile was run through Pie Charts For Everyone!, a tool created by Mohan S Nayaka. For every top author included, their name was put in one column, their gender was included in another, and a topic where the person had answered 50 or more questions was included in a third. A new line was created for each topic.

If a person had not answered 50 or more questions amongst all topics, the top contributing topic was the only one included. The decision was made to use 50 as a benchmark because it implies the ’15 top writer has been contributing for a more extended period of time in this area, and they have expertise in this area because they have answered a large number of questions in it. It is also a nice round number that should be achievable for people who knowledge specific areas they are contributing in.

One of the quirks of Pie Charts For Everyone! is that it does not include topics that authors have muted on the list. For instance, I have muted the topic Women. In going through https://www.quora.com/content?topic=3916&content_types=answers , I have answered 67 answers in the women topic. Because I have it muted, this does not appear under my list of topics for which I have include amongst other things Learning Languages,
English (language), Australia and Wikipedia but not Women or Atheism. While you can view your own mute list at https://www.quora.com/settings/muting , this information is not publicly available. It is one of the limitations of this study that people should keep in the back of their heads.

Gender was determined based on using one or more of the following criteria:

The person had a name traditionally associated with a specific gender from where their profile indicated they were from. Micheal is a traditional man’s name in the United States. Laura is a traditional woman’s name in the United States, England and Spanish speaking countries.
The person had a picture of themselves in their profile, and they appeared to be male or female.
The person said in their profile that they were male or female, or used male or female specific pronouns that indicated their gender.
The person answered a question where they indicated their gender.

The data was gathered between December 31 and January 5. The total number of ’15 top writers was 152, out of 220 that were previously identified for the research done for Understanding the ’15 Top Writer community: Network analysis . The reason why data collection stopped at this point was there was a 24 hour period where, despite multiple and repeated attempts to get new pie charts, the tool refused to work. This method of manual collecting data was already time intensive, and it was not feasible to continue with an even more time intensive methodology.

As mentioned above, only topics with 50 or more answers were included, except in cases where a writer did not have 50 or more answers in any topic. The list of ’15 top writers included in this sample who did not have 50 or more answers in a single topic included Triya Bhattacharya, Shannon Larson, Bulat Bochkariov, Christine Choi, Roderick House, Ankit Sethi, Mihika Kulkarni, andMira Zaslove. They represent 5% of the total writers included.

Another limit is that people with long contribution histories have answered in more topics than the top pie allows to be displayed as the percentages become less than 1% even as the contributor has more than 50 answers for those topics. This situation held true for at least 10 different contributors, though they were not segregated out as an assumption was made that at that point, a person’s expertise where they have answered less than 1% of their own answers in a topic, it is marginal. Therefor, there was no loss in meaning by excluding these topics.

The assumption for all data is that the number of ’15 top writers included reflects a representative sample as it includes 69.09% of all identified ’15 top writers. It should be big enough to get a general idea as the scope of topics included, and that the male and female distribution is also equal.

Results

There are more male than female writers, with 67.7% (103 total) being male and 32.2% (49 total) being female.

The topic list was quite large, with ’15 top writers having 50 or more answers in 408 different topics. Women contributed to 158 unique topics, while men contributed to 344 unique topics. Even accounting for the proportional differences, men are more likely to contribute to more subjects than women who appear to specialize. Individually, the average male contributor had expertise in 8.85 topics compared to 7.1 for women.[1]

What topics are the most popular amongst Quora ’15 top writers? There were 14 topics where 10% or more of all writers had 50 or more answers. The most popular was Quora Community, which 23% of all writers contributed to.

Some of these popular topics were more likely, in the sense of proportional representation on Quora, to be answered by women than men. This includesLife, Quora, Dating and Relationships, and Psychology. Men were much more likely to be experts, pproportionally in Politics, U.S. Politics, History,Philosophy, Economics and Religion.

Thomas Foster’s answer to What are the most followed topics on Quora?contains a list of some of the most popular topics on Quora. The list is 25 items long. Crossover categories include Movies, Food, Economics, Psychology, History, Dating and Relationships, and Politics. 50% of the topics on the most popular ’15 top writers topics appear on the most popular list. It isn’t an exact match, so ’15 top writers most common list isn’t necessarily representative of what people are interested in.

The first item appearing on most popular Quora topic list is Movies, which is 4th on the most popular on Quora but ranks 7th amongst top writers. Psychology is 7th amongst popular Quora topics but 11th amongst top writers. Economics is 10th amongst Quora contributors and 13th amongst top writers. Psychology is 11th amongst Quora contributors interest but is much higher at 6th place for ‘top writers expertise. History is 12th amongst Quora contributor interest but ranks much higher with expertise at 10th. Dating and relationships is at 17th amongst topic interest on Quora but is one of the areas where Quora top writers have the most expertise, ranking 4th. Philosophy ranks 14th amongst Quora topic interest, and 12th for top writer expertise. Politics ranks 18th in Quora topic follow interest, but ranks 5th in top writer expertise.

What are the most popular topics amongst Quora’s top female writers? The list of topics that have 10% or more of total female respondents is 17 topics long.

While there is a fair amount of crossover with the general list, some topics are unique to this list, including Women, Books, Love, Medicine and Healthcare,Parenting, Human Behavior, LGBTQ and Quora. A number of these topics generically fit with the stereotype of women’s interests and women’s concerns. Only one of these unique to women items appears on the list of the most popular topics to follow on Quora. A number of these topics are also dominated by women, including five that have 50% or more women who are topic specialists. Four of these topics have less female participation than the average representation on Quora.

What are the most popular topics amongst Quora’s top male writers? The list of topics that have 10% or more of total female respondents is 19 topics long.

Items unique to men from the general list include Atheism, Computer Programming, Religion, Music, Startups, Science and The United States of America. With the exception of possibly Atheism and The United States of America, you could argue that these topics are also very general and Quora has a fair number of male generalists.

Some of these topics would be considered more stereotypically male than female, including Computer Science, Science and Startups. Two of these items, Computer Science and Startups, would fit into that same category as generally stereotyped as being of more interest for men than women.

Who are some of the leaders of Quora’s generalists from the point of view that the top topics on the whole are very generally? [2] 115 writers out of 152 included have 50 or more answers in one of these 14 topics. That’s 75% of the ’15 top writer class. Dan Holliday leads them all with being the broadest generalist in this context. He has 9 topics where he has written 50 or more answers in that topic. Kris Rosvold, Stephanie Vardavas and Ariel Williams are close behind with 50 or more answers in 8 topics. Rounding out the top ten for generalists are Will Wister, Doug Dingus, Jesse Lashley, Jan Leadbetter,Tom Byron and Jayesh Lalwani. Women make up 20% of the top twenty, below average for their inclusion on the ’15 top writer overall number. 42, or 36% of all people appearing in the list as having answered popular generalist topics, only have one topic on that list.

What is the scope of specialist knowledge, as expressed by only one ’15 top writer having 50+ answers in a topic? That list is 241 topics long and includes some[3] of the following: Accounting, Airlines, Babies, Norway, Crime, Obamacare (Patient Protection and Affordable Care Act), Plants, Race and Ethnicity, Quora Feature Requests, Canada, Gardening, Recipes, Shoes, Sports, The Beatles (band), The Hobbit (1937 book) and The Bible. Some of these topics are very broad and some of them are much more specific. 184 of these topics appeared because a man had written 50 or more answers, while 57 appeared because a woman wrote 50 or more answers in that topic.

18 of these topics could be described as locations, including Beijing, Canada, Countries, Greece, Harvard University, Iran, Israel, Los Angeles, Middle East, Norway, Republic of Ireland, Russia, Saudi Arabia, Soviet Union, Taiwan, The Netherlands, Vancouver, BC, and Yale University. 16 appeared because of men and 2 because of female.

Specialist knowledge of location in generally does not appear to represent a large category of very specific specialism. The list of places that appear is very limited. If you expand to include all place locations like those, you get the impression that overall, top writers are not hugely into writing place specific generalist information about a location. When this list is stripped back to include only existing countries, states, and cities that are not defined as shorthand for an industry and includes topics with 1 or more top writer with 50 plus answers[4], the following picture emerges:

There are 46 unique writers who have 50 or more answers in these 26 location topics. Amongst these, only a few have more than one location topic they specialize int. These include Rupert Baines who leads with 4, Paul Denlinger,Andy Lee Chaisiri and Andrew Gutsch with three, and Viola Yee, Joe Geronimo Martinez, David Stewart and Jeremy Miles with two.

This list and relatively popularity of ’15 top writer specialization for writing about location does not necessarily match with Quora’s readership. According to quora.com Site Overview, India accounts for 38.9% of all Quora traffic while the United States for 29.9% of all traffic, the United Kingdom for 2.3%, Canada for 2.1%, and China for 1.5%. In this regards, the ’15 top writers cohort specialization may not match with reader demographics.

How diffuse and connected is Quora’s knowledge expertise? A network analysis was done for this, and the following graph was generated.[5]

This network has 3 connected compontents, with a network radius of 1 and a network centralization of 0.055. The characteristic path length is 4.382 with the average number of neighbors at 4.508. All told, there are 559 nodes, 1 self-loop and 2 multi-edge node pairs.

Like Understanding the ’15 Top Writer community: Network analysis , there were isolated pairs which suggest that despite the illusion that top writers are a clique, they are not in the sense that they are not all writing in the same spaces on the same topic. Some contributors are isolated from others in their writing interests. This includes authors like Kåre Lohse, who connects to the network as shown in the graph below.

Most users are a bit more connected. Some general knowledge area specialists who have many answers in specific topics around that general area appear like flowers, connected to the whole by one or two topics, but having lots of individual topics where they are alone in their expertise.

You can see this above for David Joyce who specializes in physics and mathematics related topics, Isaac Gaetz who specializes in engineering topics and Christopher VanLang who specializes in biology topics. While not visible, it is also try for ’15 top writers like Mark Hughes who specializes in comic related topics, Thomas Snerdley who specializes in Lord of the Rings, Dorin Lazăr who specializes in computer programming, and William Chen who specializes in data.

The graph above featuring Elynn Lee and Gayle Laakmann McDowell gives a clearer picture as to how specific writers connect topically to each other to form this network of content knowledge on Quora. These authors are largely specialists in their own domains, with the exception of one shared one.

Topics also sometimes appear the same as writers, looking like their own little flowers of sorts. In this case, the graphic above shows ’15 top writers who have written extensively about Indiaincluding Shefaly Yogendra, Sireesha Chilakamarri, Siddhartha Das, Sudhendu Pandey and Balaji Viswanathan. The last two are networked more into the broader Quora environment. As discussed in the previous post looking at the relationships between top writers, this idea does to a degree confirm to a degree that there is a culture gulf between Quora users from the west and Quora users from India.

Conclusions

Quora’s ’15 top writer class has depth in subject area expertise and depth in the number of topics they can answer, with over 400 different subjects they can and have provided many answers in. There is a decent cohort of generalists who can and do answer general topics, though their expertise does not necessarily align directly with the most popular and most followed topics on Quora.

There is also a cohort of writers that write about multiple but similar topics, serving as both generalists and specialists for these topics. These specialist nodes run the gamut from popular culture, science, technology, specific geographic areas to careers.

This data also reinforces the existence of Quora’s gendergap in terms of male/female participation. There are more men, and they actively engage more across more topic areas. There is limited evidence to suggest that gender expertise of the cohort of ’15 top writers mirrors stereotyped gendered expressed interests on other websites.

The geography specialty distribution does not necessarily reflect Quora’s readership. In the long term, Quora may want to consider trying to improve specialization to better reflect and engage readership by encouraging people who reflect it.

In doing a network analysis, the conclusion that there are two Quoras with a western culture Quora and an Indian Quora sounds correct. ’15 top writers about India are largely experts in India and are not very topic networked with other writers who appear to dominate the ranks of ’15 top writers.

It would be interesting to explore this topic further and better understand the relationships between various topics, and the writers who serve as links between them. It would also be interesting to explore in more depth the concept of generalist versus specialists, people who answer questions across multiple broad topics versus people who answer questions across multiple topics in the same area. How does their behavior change in terms of their other interactions on Quora? Does this impact how others treat them in terms of upvotes and follows?

Footnotes

This number was derived by taking the total number of topics next to rows that had a specific gender in them and dividing that total by the number of contributors for that gender. It was not derived by dividing based on total unique topics.
This number was derived by filtering people only based on topics from the 14 most popular topic list: Dating and Relationships, Economics, Food, History, Life, Movies, Philosophy, Politics, Psychology, Quora, Quora Community, Religion, Survey Questions, U.S. Politics. Then the number of times each unique name matching that topic appearance was counted. It is worth remembering that not all top writers appeared in this study because of limitations with the data collection tool. It represents only around 70% of all top writers, and a lot of people were left out. This list is not conclusive.
The complete list is 2012 Summer Olympics in London, 3D Printing, Accounting, Airlines, Airplanes, Algebra, Apple Products & Service, Arab Spring: Protests & Revolution in the Middle East (2011), Architecture, Art History, Asian Americans, Asian People, Astrology, Autism Spectrum, Babies, Batman (comics, movie and creative franchise), Beijing, Biochemistry, Biotechnology, Birds, Blogs, Business in China, Business Strategy, C (programming language), C# (programming language), C++ (programming language), Calculus, Canada, Chinese (language), Civil Engineering, Comic Books, Communist Party of China, Computer Programmers, Conflicts, Consciousness, Conservatism (politics), Construction, Contemporary Art, Contemporary Classical Music, Cosmology, Countries, Creative Writing, Crime, Crowdfunding, Culture (social science), Data Analysis, Data Science, DC Comics, Definitive Albums, Democracy, Democratic Party (U.S.), Doctor Who (TV series), Doctors, Dogs (pets), Domestic Violence and Abuse, Drug and Pharmaceutical Companies, Ecology, Electrical Engineering, Electronic Music, Electronics, Emotions, English as a Foreign Language, Evolution (process), Exercise, Fabrics and Textiles, Facebook (product), Family Relationships and Dynamics, Flowers, Functions (mathematics), Game Design, Game of Thrones (TV series, Gardening, Gayle Laakmann McDowell (author), General Relativity, Genetics and Heredity, Geometry, Google Internships, Government, Graduate School Admissions, Greece, Gun Control, Harvard University, Health Insurance, Higgs Boson, Hinduism, Hip Hop Music, Historical India, History of China, Hong Kong, Hospitals, Human Brain, Humor, Immigration, Infectious Diseases, International Travel, Internet, Internet Forums, Internships, iOS, iOS Development, iPad, iPhone Applications, Iran, Islam, Israel, J. R. R. Tolkien, Japanese (language), JavaScript (programming language), Jewish Culture, Jewish Ethnicity and People, Job Interviews, Laptops, Lean Startups, Learning, Learning English, Learning to Program, Lesbians, LGBTQ Issue, LGBTQ Rights, Life Experiences, Linux, Logic (philosophy), Los Angeles, Mac (computer), Machine Learning, Manga, Marc Bodnick, Marvel Comics, Materials Science and Engineering, Mechanical Engineering, Medical Conditions and Diseases, Mental Illness, Middle East, Mobile Phones, Modern Art, Molecular Biology, Motherhood, Movie Production, Movie Recommendations, Music Composition, Musicians, National Basketball Association (NBA), Norway, Novels, Nudity, Number Theory, Numbers, Nutrition, Obamacare (Patient Protection and Affordable Care Act), Objective-C (programming language), Olympic Games, OS X, Parenting in Education, PC Gaming, People of China, Pianos, Plant Identification, Plants, Poetry, Pokémon, Police and Law Enforcement, Politics of the United Kingdom, Polyamory, Premiere League, Probability (statistics), Programming Interviews, Proofs (mathematics and science concept), Pyschology of Everyday Life, Quantum Field Theory, Quora Feature Requests, Race and Ethnicity, Reading, Recipes, Relationship Advice, Relativity (physics), Republic of Ireland, Resumes and CVs, Russia, Saudi Arabia, Seed Funding, Self-Improvement, Semiconductors, Senior Citizens, Shoes, Sleep, Small Businesses, Smartphones, Social Media, Social Media Marketing, Songs, Soviet Union, Spanish (language), Special Relativity, Specific Social Networks (online), Spiders, Sports, Startups in New York City, Stock Market, Structural Engineering, Superheroes, Surgery, Taiwan, Teachers, Teaching, Technology Investing, Teenagers and Teenage Years, The Beatles (band), The Bible, The Church of Jesus Christ of Latter-day Saints, The College and University Experience, The Future, The Hobbit (1937 book), The Lord of the Rings (books, movies, and creative franchise), The Netherlands, The Silmarillion (1977 Book), The Universe, Theater, Tips and Hacks for Everyday Life, Top Writers (Quora program), Transgender, Trees (plants), Trigonometry (mathematics), U.S. Congress, U.S. Constitution, U.S. Department of State, U.S. Presidents, United States Armed Forces, Vancouver, BC, Veganism, Vegetarianism, Venture and Investor Pitches, Venture Capital Firms, Video Game Industry, Web Development, Wikipedia, William Shakespeare, Wireless Technology, Word Definitions, Terminology, and Jargon, World War II, Writing Advice, Yale University.
This number was derived by filtering people only based on locations that were cities, states or countries. Then the number of times each unique name matching that topic appearance was counted. It is worth remembering that not all top writers appeared in this study because of limitations with the data collection tool. It represents only around 70% of all top writers, and a lot of people were left out. This list is not conclusive.
This graph was generated using Cytoscape. The data for this included a table with two columns. One column included the user name and the other column included a topic where the author had 50 or more answers in that topic.
This topic could have easily been explored by using writers other than top writers. Top writers gives a finite and defined community. It is a workable population. It isn’t random, so it clearly is not going to be representative. It gives insight into a specific group of writers who have recognized high quality content and who are active in the community. It is also generally safe to assume that most of the writers in this group are likely to be specialist in some area, and have 50 or more answers in a single topic.

Written 6 Jan.

Understanding the ’15 Top Writer community: Network analysis

May 17, 2015 Laura Hale quora

Disclaimer: I do not normally do network and node analysis. Apologies. This is quasi-scientific in the sense that it is written generally like an academic paper, minus the references and overly formalist tone. Hopefully, this makes sense.

This was not as anonymized as well it probably could be on the belief that a lot of Quora is built on reputation from answering questions. If you see a name that you think should be anonymized or want your name removed, suggest the edit and it will be removed. All of the data used to build this was publicly available on Quora.

Introduction

There are a lot of Quora questions that ask questions to top writers, complain about how people who are not top writers do not get readers and upvoters, and imply there is a clique that can have negative consequences for those outside of it through lack of upvotes, views and new followers.

Earlier, I explored this with Being a top writer means I am in an awesome upvoting clique! Right? I mean, really? After some more pondering on the subject, I felt like a slightly deeper analysis might be useful in terms of understanding the dynamics at play on Quora when it comes to ’15 top writer upvote interactions only inside the group. The insights from this analysis might have other benefits in terms of understanding some of the broader community interactions on Quora, including areas like what content you see on your feed.

This analysis is looking to better understand the relationships between ’15 Top Writers, and how they are socially connected to each other through their upvoting behavior. This is being done to understand this group of writers, which have been at times described as a super clique who have a negative impact on Quora by feeding into a hive mind of content. The analysis relies on a node analysis, looking at matched pairs of ’15 Top Writer upvoting an answer by another top writer/’15 Top Writer having been upvoted by a top writer.

This is not designed to be the “be all, end all” analysis to Quora analysis because there are a lot of limitations in data collection. Hopefully, people interested in knowing more how their community works or interested in social media communities will find it interesting. The purpose at the end of the day is to get insight to do just that.

Methodology
A list of ’15 Top Writers upvotes for other ’15 Top Writers was manually gathered by looking at different questions. When a top vote pair vote was found, it was recorded. Only votes visible from the answer itself without needing to be clicked through to “View All” were recorded because the hover option was unavailable from that location. This is a limitation of the data collection because upvote pairs deep in massively upvoted answers were not counted. All answers, and their ’15 Top Writers/ ’15 Top Writers pairings, to any question looked at were recorded.

The image below gives an idea as to how this data was collected on Quora, relying on the little red symbol to easily identify this year’s top writers.

This information was then collected on a spreadsheet between December 25 and December 28. With the example above, it would have been recorded with Flom in a column marked Upvoting, and Miles marked in the column marked Upvoter.

Questions were initially chosen based on looking for geographically distinct topics, and then picking questions at random in order to try to get a broad representation of different Quora topical content. After this, questions were chosen that appeared on my content feed. At this point, the total number of ’15 Top Writers included about 220. A deliberate effort was then made to try to get more ’15 Top Writers represented, with Top Writers 2015: Which Quorans are Top Writers for the first time this year (2015 – announced November 2014)?and other posts with lists of writers under the Top Writers 2015 topic as a place to find writers not on the list and include them more by finding questions they had either answered or upvoted. At that point, the list was 281 unique ’15 Top Writers. I cannot find a complete list to know how many people are actually on the list, but a previous year had around 440 top writers to it seems fair to assume this list includes probably about 2/3rds to 3/4ths of all top writers.

Collapsed answers were always checked on questions to make sure top writers had upvoted answers that were not being credited. This happened only on two occasions. One was collapsed because of downvotes, but had upvotes from other top writers. The other was collapsed because it was not written in English, and had no upvotes from other top writers.

The list of matched pairs draws from 1,095 upvotes for 266 answers from 176 different questions. It is worth noting that not all ’15 Top Writer answers looked at had visible upvotes from from other ’15 Top Writers. There were 90 occurrences where either a top writer had 0 upvotes, had upvotes but not from other top writers, or possibly had upvotes from other top writers but they were buried deep on highly upvoted answers. About 25% of the time, top writers were not guaranteed to have upvotes. These answers ranged from ones by writers with several thousand followers to those with a few hundred.

In trying to find answers by top writers, there were some difficulties that can explain the large total number of answers by top writers without other top writer upvotes. The first is that the question and answers were too new to draw too much attention. The second is that the answers were very old and predated the person becoming a top writer. The third is that the topics were very niche, without what appeared to be too many other top writers with an interest in it. The last reason, discussed in more depth in the section titled The Other Quora, is the question related to India where few top writers appeared to be getting upvotes from other top writers, even when they were both writing on the same topic.

It is worth noting at this point that this list is not representative of anything except some observed relationships with the belief that these relationships would likely constitute a representative sample in terms of understanding relationship dynamics. The total volume of questions, total volume of upvotes, total number of top writers, total number of upvotes each gave to other top writers is all unknown as these numbers are nor publicly available, nor easily publicly derivable without the use of a bot or API. The latter are not available options due to lack of a Quora API.

Results

Of the 281 ‘Top Writers picked up in this data collection, the average total mention per person was 7.8 times, but there clearly was a pattern evident in the mode that people were much more likely to be upvoting than they were to be the upvoted. There was not a 1:1 relationship, with a person getting an upvote for every time they upvoted.

There were 108 ’15 Top Writers who were recorded as upvoting an answer but not having an answer that was upvoted. This contrasts with 60 people where the inverse was true.

This more likely to upvote than get upvoted matters a lot, because some of the people who upvote more than get upvotes have large followings. The top upvoter found in this group has 16,200 followers, with 53 upvotes to 7 times having been upvoted. The second person on the list of most upvotes has 1,500 followers, and recorded 33 upvotes while having been upvoted 13 times. The third most prolific upvoter has 8,800 followers, and recorded 30 upvotes while having been upvoted 0 times. The fourth person had 5,900 followers, while having recorded 28 upvotes and 0 answers being upvoted. There was a three way tie for fifth with 26 upvotes. One has 3,100 followers, while having 26 upvotes, and 13 of their answers upvoted. The other has 2,700 followers, while having 26 upvotes and having their answers upvoted 11 times. The last person has 700 followers, with 26 upvotes while having their own content upvoted 0 times.

The top ’15 Top Writer upvoted had 34 upvoted answers while only upvoting 22 times. With 3,400 followers, they also serve as a crucial node in Quora content visibility. The second most upvoted writer had 33 upvotes while upvoting 0 times, and having a whopping 32,800 followers. In third is a user with 3,800 followers who had their answers upvoted 30 times while only upvoting other people’s answers 4 times. In fourth was a user who had their answers upvoted 29 times while upvoting others 6 times, and having 3,300 followers. In fifth is a user with 25,700 followers who had 27 top writers upvoting their content while upvoting 10 other top writer answers.

These seven upvoters and five upvoted all have the potential to bring tremendous influence to their audience because they are central nodes in content generation for their follower’s home page and amongst top writers. They are part of the central hub on Quora, of which there found to be three. One has 278 writers, the second has 3 writers and the third has 2 writers.

This is visualized as follows when mapped for closeness for size and betweeness for size.

A close up view of this network looks like the following:

Some of the relationships are stronger than others, with there being a frequency in people upvoting certain writers though this is not always bidirectional. There were 49 pairs picked up where a writer upvoted another writer more than once. When mapped out and including their other non-multiple upvotes, you get a network that looks like the following:

There are a few dead ends here, but there are some very, very networked writers who are connected to other authors. There are no isolated networks here. Many of the names found here are ones who rank as key nodes for upvoting or having been upvoted.

These two views confirm that Quora users are networked with others, and to a certain degree confirms there is a clique though the deliberateness of this clique is up for debate because there is no real strength in these answer volume. This could be the result of not having many writers with more than 1 answer appearing. Of the 266 total answers, they were written by 173 unique writers, with only 59 writers appearing more than once with their answers.

What is less obvious in the graphs is the connectors, and there are a few important connectors hidden here that become more obvious when you start pulling users out and looking at betweeness centrality. These connectors sometimes often have their own high volume of followers, but also serve as bridges between different topical intersections where two nodes connect. These topical connectors include people like Jennifer Apacible, Christopher J. Su,Mark Barton, David Stewart and Ankit Sethi.

The diversity of the network can be difficult. The distance between some people can be pretty great. (In some cases, it does not exist.) For example, the path between Deepak Mehta and Martin Fox requires going through a number of people including Shubham Choudhary and then Lorenzo Peroniwho plugs into the main network, which will get you to one of the three people going to Nathan Ketsdever who can then finally get you to Fox. These two users may be on the same site, but their close networks and interests are generally vastly different. The average shortest path is about eight users. That concept of clique is hard to maintain when the shortest path is that big.

The Other Quora
If you read through some commentary and criticism of Quora, the topic of India often comes up. People sometimes talk of either feeling like Quora is a site for Indians that others happen to use, or of two Quoras existing on the same site.

A notable example of this two Quora phenomena can be see with Rohit Nigam, an Indian ’15 top writer. He has 624 followers, which is a decent number and comparable to other members of the class of ’15. Being able to include him in the node analysis was difficult and very deliberate. After looking through his answer page, the first 20 answers did not have a single upvote from a top writer. At the same time, while Rohit upvotes a lot of content, he generally is not upvoting content of other ’15 top writers. He participates almost entirely on The Other Quora. He was not alone in this, with a number of other ’15 Top Writers from India and Asia having similar interactions, or in this case lack of interactions through upvoting, with their fellow top writers. This lack of interaction with other top writers often extends to upvoting their answers on questions they have answered. One example of this is If you were to write the last line of the last tome of A Song of Ice and Fire, what would it be? , where there are at least three answers by ’15 top writers, none of whom were early on the list of upvoting each other’s answers.

The Other Quora phenomena can be magnified in who you follow. There is a cohort of high ’15 Top Writers who have a lot of followers and who upvote content of other ’15 Top Writers. Once inside this particular group, your Quora content will often reflect this particular node unless you’re making an effort to follow across it by following more popular, not ’15 Top Writers from The Other Quora. There is a similar thing going on inside The Other Quora, with many Indian contributors following fellow Indians, thus creating a different network that is reflected in the content they see and answer, and thus feed into other people’s viewing experiences.

The idea of the existence of The Other Quora is fully supported when you start examining network diameter and connectors.

Conclusion
Quora’s top writers are part of a community. Many are isolated several steps away from other users, and a few are almost completely isolated from the rest because of a lack of connecting users. The large number of connectors mean that Quora users can exist far away from each other.

While the analysis shows the existence of super users who broadly speaking can have a tremendous impact on content fed to other users, this is not a self perpetuating cycle of only top writers because of the broadness and diameter of the ’15 Top Writer network interconnectivity. The concept of a clique is only supported inside a small number of highly connected top writers who actively participate in the site.

For future research, the data could be expanded to include more ’15 Top Writers and to include more upvoting/upvoted pairs. Not all writers were included, and the lack of more questions and repeats on author questions makes the results somewhat limited as there is limited insight into the strength of some of the demonstrated relationships.

It may also be useful in terms of delving deeper into the topic of The Other Quora to better understand the differences in how Western and Indian users interact with each other and Quora. It would also be worthwhile to investigate the socialness of self-identified male versus female ’15 top writers and if their behavior and networks differ from one another given Quora’s gendergap.

Appendix 1: Questions included
The following is the list of questions from which upvoting/upvoter pairs were culled.

* Why do people enjoy learning and making conlangs (artificial languages)?
* Did Bertrand Russell really say, “I have nothing to hang onto but grim, unyielding despair”?
* If there are no limits to the number of topics one may follow on Quora why am I unable to follow any topics during the last 24 hours?
* Why has the population of Chicago been declining for the last 60 years?
* Do women like unintelligent men?
* Which startups have the best stories?
* Will men always be men?
* What is the coolest looking animal?
* Are Top Writers a clique?
* Are there cruel animals?
* What are the best Japanese anime movies?
* In America, if hamburgers are actually made of beef, then what do they call burgers made of ham?
* What’s the difference between ham and pork?
* How do the various peoples of Micronesia (Federated States of Micronesia, Marshall Islands, Palau, Mariana Islands) feel towards Japan and its previous colonisation?
* What can an individual or a family or a business in the Northern Marianas Islands do today to prepare and adapt for the coming impacts of climate change?
* Why wasn’t there a sequel to Top Gun?
* Why don’t US citizens who are proponents of Socialism and tout European Socialist ideals immigrate to those countries?
* Why do some people observe everything I do?
* Why is Israel so big in high tech?
* Has there ever been, in all of history, any time and place where Jews were safe from persecution?
* Which is worse: bulimia or anorexia?
* Is it possible that everyone is lying to me and I really am fat?
* I really need help with my bulimia, what should I do?
* Why do most Russian last names end with “ev”, “ov”, “sky”, or “in”?
* Why do Russians use Cyrillic instead of the usual alphabet?
* What is the optimal age for a child to learn a new language?
* How do I upvote answers anonymously on the Quora mobile app?
* Is it legal to set traps to catch wild rabbits in California, or do I need a license?
* Is there research behind why people seem to get hit on more when they’re in a relationship?
* Should I continue to work or apply for an MBA?
* How is the sound of a laugh written out in different languages?
* Christians and other theists: When you have a discussion about religion and atheism (outside of Quora) with an atheist, who usually brings up the topic?
* Why don’t people learn from their mistakes? I still continue to take people at face value.
* Does “the present” actually exist in Doctor Who?
* Is it extremely unusual for a person to type/write/text in better English when they write English, as opposed to when one is speaking face to face with somebody in English?
* What was bowling like before automation?
* What do you think of this mini documentary on automation by CGPGrey?
* What would we do if someday all jobs are automated?
* What is the most frightening thing most people don’t know about the US government?
* Which is easier to learn, Spanish or Portuguese?
* What do atheists think about the Argument from religious experiences for the existence of God?
* Why is Africa so poor?
* What are possible strategies for developing Africa?
* Is China colonizing Africa?
* What are the behaviors and attitudes of Indian women that the men find difficult to contend with?
* Why can’t Indian women say “no” (to relationship approaches)?
* What is the biggest misconception about Indian women?
* Bangladesh: Do Bangladeshis hate Pakistanis?
* What is going on in Bangladesh?
* Will Bangladesh ever become the ‘Islamic Republic of Bangladesh’?
* What are some of the most iconic images from cricket?
* What’s the difference between baseball and cricket?
* Why do many soccer players not make it to the top level?
* Why would the IDF arrest Palestinian teens playing soccer?
* Is abortion morally right?
* Abortion Rights Movement: When do pro-choice adherents believe life begins?
* Why is it that entrepreneurs lie about their motivations, pretending that money is not one of their goals at all?
* Why do Cambodians have so little general knowledge?
* What do topic pie-charts for Top Writers look like?
* Which state in Australia is the most inclusive city for people with disabilities?
* Mimi Metcalfe’s answer to Is Australia a good place for immigration? Do you intend to immigrate to Australia?
* Do music and mathematics have common “parents”?
* How do I “quit” someone?
* Was preparing for the IIT-JEE worth it?
* Are there squirrels in Germany?
* What is Germany bad at?
* Will Germany remilitarize?
* Should other nations follow Germany’s lead on promoting solar power?
* How do Duolingo levels work?
* What CEFR level is Duolingo German?
* What is Putin’s endgame in Ukraine?
* Is Ukraine’s Communist Party now illegal in Ukraine?
* Should I write “Ukraine” or “the Ukraine”?
* What is it like to live in Lviv, Ukraine?
* What should America’s gun control policy look like?
* My 15 year old daughter is identifying as lesbian to friends, but hasn’t told me. What is some good advice on how I should approach her?
* Is Hindi a dying language?
* What are best Hindi or Indian jokes out there?
* Is Obama good at Hindi?
* Could India and Pakistan unite again?
* What is the “Idea of Pakistan”?
* Why is media mostly behind negativity?
* Social Media Analytics: Which data is most important to your brand in regards to gauging the effectiveness of content?
* Is Wikipedia going commercial?
* How commercialized is Wikipedia?
* How is Quora different from Wikipedia?
* How can the social media platform Ello create monetary value without adding advertisements?
* Do music and mathematics have common “parents”?
* What is the most efficient way to become a Quora millionaire?
* There is plenty of circumstantial evidence of police racism, but is there any scientific data proving that it’s an issue?
* What scientific evidence proves that Bigfoot could exist?
* What scientific and historical evidence proves the existence of Noah’s ark?
* If I post on my Facebook Timeline, does it come up on my friends’ pages? Or do they have to visit my page to see it?
* What is the difference between an affidavit and a testimony?
* How do you respond when a girl says,”You’re a geek”?
* My wife’s ex-boyfriend’s child has the same name as the company she is working for. Should I be pissed?
* Which is the best country in which to learn Spanish?
* How many Top Writers 2015 are teenagers?
* Vivek Mehta’s answer to How do I know what I should use between can and could?
* I need a summer internship, but I don’t want to apply because there’s a 90% chance I’ll get rejected. What should I do?
* Were Sundar Pichai and Andy Rubin enemies inside Google?
* If all Google engineers had to re-interview to keep their job, what percentage would succeed?
* How would a new grad with no experience in product management prepare for a PM interview?
* Why do we get so squeamish when we hear a chalkboard get scratched?
* Can infinite sums and integrals be swapped anytime?
* Sentences: What’s a true statement that is not helpful?
* Should the ACM programming contest be banned because it incites racist attitudes by its finals rankings?
* What is the one piece of life advice you wish someone could have told you when you were younger?
* What are some of the best answers the ‘Top Writers 2015’ have written?
* Is it accurate to say “yellow fever” is white men liking or being romantically attracted to East Asian women? Or is such a term just restrictive terminology for a type of fetish aimed at certain broad groups? Why? Why not?
* What are some tips for programming on whiteboards?
* Why do astronauts need to shower in space?
* What makes women message men on dating websites?
* How can I politely explain why I don’t want to have sex with someone who is 30+ years older than me in a way that makes sense?
* Why do I hate religion?
* What are the best feminist blogs or news sites?
* What is it like to have your own parent as your class teacher in school?
* Apart from politics, what are the best and worst things that happened to India in 2014 ?
* How has your 2014 been?
* What are the most awaited movies of 2015?
* What do women think about men who cry in front of them?
* Can you find the next number in the sequence “2, 8, 12, 8, 16, 24, …”? What is the pattern here?
* Was the megalodon the top predator of its time? Did megalodon have any predators?
* Can newborn babies distinguish colors?
* How cold is outer space? Would someone die if they were to be bare-skinned in outer space?
* Indian Ethnicity and People: As an Indian girl or woman, what would be the one piece of advice you’d want to give to an Indian boy/man?
* What are some mythological things that represent knowledge or prescience?
* Ancient Egypt was conquered by Persians and Assyrians. How come all these kingdoms preserved the Ancient artifacts like pyramids, Egyptian God idols, and other infrastructure? What has made the Egyptian pyramids stand the test of time?
* Why is North Korea joining the Sony hacking probe? Why are they suddenly scared of the US?
* What are some famous telescope accidents?
* Is life in North Korea different from what we have been told?
* What is the deepest thinking dog breed?
* How did the US actually force a denial of service attack against North Korea without impacting or compromising China, who hosts most of the infrastructure North Korea relies upon?
* Is Britain under threat from Islamists, considering the wave of sex crimes committed by Muslims targeting young white girls?
* What type of relationships do individual Russians and Americans have inside of their respective countries, and elsewhere?
* Does it benefit anybody to discuss social justice issues on Quora? Is it worth doing?
* Other than touring wineries, what can you do in Napa/Sonoma?
* Did a cop really have a weapon accidentally discharge in a classroom full of little children?
* Where in Australia should I worry about racism?
* Would you marry a woman who would sleep with you on the very first date?
* Delhi election surveys of news channels are showing Arvind Kejriwal is the number 1 choice for CM, showing that a majority of people (more than 65%) were happy with AAP’s work but they are showing that the BJP will get the majority. Are they not contradicting themselves or are they trying to make a fool of the people?
* Sanskrit replacing German in schools, the proposal to make the Bhagavad Gita the national book of India, Sadhvi’s ugly utterances, and the beatification of Gandhi’s murderer. How do Quorans who support the BJP view all this?
* Why does “dharna” or peaceful protest have a negative connotation in India knowing the fact that it is one of the basic features of a democratic system?
* What are the reasons behind the Modi government’s decision to cut the health budget by about 20%?
* Has there been any significant development in Rajasthan under BJP’s government?
* What is your opinion on “spare the rod, spoil the child”?
* What is worse: being lazy or being dumb?
* Is drinking Coke Zero bad for your health, particularly with regards to weight gain?
* Will the AAP support the bill to stop forceful or lured conversion?
* Will the “Cyber war” between BJP and AAP give time to INC for reorganising themselves?
* Why do the current advertisements from the AAP on the radio say ‘Currently the BJP is running the government in Delhi’? How is that happening?
* Do you expect India to have 100 smart cities when Modi has not been able to give even one smart city in 12 years term of Gujarat?
* Why did IIT Delhi director resign?
* Stupidity: What are the benefits of being dumb?
* Would I truly mess my children up beyond repair if I left them and their father?
* How would cross posting on Quora Blogs and WordPress affect SEO?
* Why do most Indians want do their master’s abroad?
* While Atal Bihari Vajpayee deserves a Bharat Ratna, was this a wise decision by the BJP to come to power and give the first Bharat Ratna to its own leader?
* How can I be an official member of Wikipedia?
* Was YouTube sold too soon?
* What is that vest that Netherlands players were wearing before the extra time in quarter finals vs Costa Rica?
* If you were to write the last line of the last tome of A Song of Ice and Fire, what would it be?
* Scientists (and science enthusiasts) of Quora: What is your personal “holy grail” of science?
* I’m a sophomore in college at the moment. When would be the best time to start practicing for the MCAT?
* How much do practicing physicians use the biochemistry they learned in med school?
* How do I save the meal when I have added too much chili pepper?
* Why does Ygritte say “you know nothing, Jon Snow” when she encounters him in Castle Black?
* Can a pure mathematician as skilled as Terrence Tao easily understand theoretical-physics papers on M-theory?
* If you encounter a gorgeous, intelligent woman married to an ugly man who is not rich and does not have a college degree, what is usually the secret behind the man’s success?
* What are common dialogues you will hear from an auto rickshaw driver in India?
* Is the cost of PM Narendra Modi’s spectacles 1.8 lakh rupees?
* If Gandhi is father of the nation who is the mother?
* Why do most of the people go abroad after engineering in India? Are they frustrated about India?
* Am I allowed to sign up for the Selective Service in the US even if I am female?
* Who are the British that are still honored by Indians for their great deeds?
* When a linguist (i.e. a scientist who works in linguistics) claims to know, say, ten languages, typically, how many can he sustain an ordinary conversation in?

Written 29 Dec.

Being a top writer means I am in an awesome upvoting clique! Right? I mean, really?

May 17, 2015 Laura Hale cliques, quora, top writer

Merry Christmas to me. A little data analysis as a present while I am busy not studying Portuguese or working on my thesis.

Inspired by Are Top Writers a clique? , I thought I would see if I could get any data to look at top writer cliques in terms of upvoting. This sort of node analysis is beyond Excel in most cases, but I’ll try to make it work.

First, I went to a bunch of questions, looked for answers by a ’15 top writer. I recorded their name. I then recorded any of upvotes by ’15 top writers. To make sure I was not getting a completely self selecting sample based on my own Quora front page, I went surfing through tags I do not normally read in.

Worth noting that while I was doing this, I ran into some answers by top writers that had zero upvotes, or had zero upvotes from any top writers. This isn’t really easy to put into my dataset. I counted 13 occurrences of this in the 25 questions I looked at.

Also worth noting, top writers were often not the top answer for questions looked at, rarely dominated in the big questions for most answers, and almost always had more non-top writer upvoters than top writer upvoters. (Translation: There is no exclusionary top-writer upvote cabal.)

This analysis is purely a node analysis based on a limited upvoting sample to see if there are little cliques of upvoting that might be easily visible.

One constraint with this data, because I manually datamine, is that I only went through the visible votes on the page, supplemented when I clicked on “more”. The additional click for the full list I bypassed because hovering did not pull up the user information that quickly told me if the person was a top writer.

Onwards with data. 25 questions were included, which included 56 unique ’15 top writer answers and 249 upvotes by other ’15 top writers. 102 different top writers upvoted the answers of 48 different writers. The most prolific ’15 top writer upvoter had 15. 52 writers only upvoted one person. There were 245 unique upvoter/upvoted pairs.

One writer had 27 ’15 top writer upvotes. 7 writers only had 1 ’15 writer upvote.

When doing a node analysis of the upvoter/upvoted pairs, there were no unconnected pairs. Everyone in the network is connected to everyone else through at least one pair. There are 14 multi-edge node pairs. The average number of neighbors is 3.95. There is a network diameter of 8.

This graph isn’t great but it should give an idea that some top writers can and do attract more upvotes from their fellow top writers than others, and that some top writers are more likely to upvote than others. It should give an idea that lots of writers are participating on the fringe, likely in their own area of interests. It should also make it clear that there are a lot of top writers who go about participating on Quora just like everyone else. This sample suggests that there is no upvote clique by ’15 top writers.

Being name a top writer does not automatically lead to getting more upvotes from your peers.

Written 26 Dec.

Variables for predicting answer placement: How much does the downvote likely matter?

May 17, 2015 Laura Hale downvoting, quora

Why does Quora allow cost free, no explanation, anonymous down voting? is astandard type of question in certain parts of Quora that implicitly criticize Quora without really expressing an understanding of what the algorithm for understanding Quora answer ranking is. To be fair, this is not public because Quora wants to prevent gaming the system. It is one of them secrets like Google search ranking also is.

This sometimes leads to… paranoia about upvote and downvote brigades that hurt people’s content visibility. I’m going to take a very small crack at trying to understand somewhat how relative answer ranking works because I’d rather disabuse the notion that downvotes are killing Quora.

I’m not going to do a very large sample size. (That is really needed to have any real idea to be able to honestly crack the magic ranking code.) I’m going to go only with questions that have exactly five answers and are in the same topic. There are also seemed to be a greater likelihood of the same people answering questions, which means less time data mining wise. This is to keep manual data mining manageable and to have some control variables in place. It allows for better direct comparisons.

Please bear in mind that the number of variables are finite but at the same time pretty huge. I’m opting for more variables at the expense of a bigger sample size. This is worth keeping in mind if you’re repeating this. Lots of variables were not looked at, such as if a person was A2Aed, profile completeness, number of links in the answer, etc.

I’ve chosen the topic language learning, because I am familiar with it and it often represents a broad cross section of Quora users in terms of geography, writing skills, gender, primary language and interests.

First question I looked at Is reading newspaper good for improving English?. It has one of those “Look! 2 upvotes is ranked higher than 3 upvotes!”

What variables could be at play? Why might this be?

The first thing that we can see is that the more highly positioned response is longer, and has more views, was written by a person who has more answers, who has more followers and has a much different followers:following ratio.

These answers and people answering aren’t comparable. The data suggests to me that the placement makes sense. A single downvote is unlikely to have accounted for the difference in placement.

This is Should I learn French or Spanish?. Again, we see that pattern where someone with more upvotes is ranked below someone with more. We actually see it twice.

This is one of those things where observational analysis doesn’t really help much because potential for confirmation bias. I’m still going to do it anyway before doing some very basic statistics at the end.

Older answers could be getting a benefit here. People with more total answers could be getting a benefit here. (Which is actually what happened in the first sample.) We could be seeing an issue of the follower ratio playing a role, those following more people than they follow get dinged, though that only works in one case here.

This one bucks the trend set by the first two: The rankings pretty much mirror the upvoting. There are two sets of ties. What is the difference for why one is ranked higher than the other?

It doesn’t appear to be total answers, because we have split differences. It does not appear to be word count because split difference. It does not appear to be the ratio of followers:following as that is split. It could be the older answer gets preference. It could also be the answer with fewer views is the one that is higher ranked, and also be the smaller difference between views:upvotes is higher ranked.

Why do people enjoy learning and making conlangs (artificial languages)? is the fourth question. This one sees two occurrences of more upvoted answers appearing lower.

If looking at them as matched pairs, it looks like the most recent answers appeared at the top. It also looks like longer answers were lower. It also looks like answers with more views were preferenced higher. (This pattern holds true in both sets: Longer response, fewer views.) More followers looks like means fewer upvotes puts you ahead in both cases. It also looks like following more has an advantage.

Is there any story or novel that uses most of the English vocabulary? is the fifth and final question looked at. The answer with the greatest total number of upvotes is ranked last. The answer with the second most number of upvotes is positioned second, but behind an answer with one fewer upvotes.

In the first pair, it looks like the person who answered more was positioned higher. It also looks like the person with more followers and following more was positioned higher. They also had a longer answer and more views. Their answer was also more recent.

In the case of the second pairing, for rankings 4 and 5, the inverse appears to be true. The newer answer is lower. More views is ranked lower. More followers, following and answers is ranked lower. Longer is positioned higher though.

Done with that look at five answers. This casual inference suggests things are totally random. Let’s total some of these positions to see if what the patterns actually appear to be when looked at together.

In these 5 answers, there were nine sets of an answer with fewer upvotes being positioned above an answer with more upvotes.

If I was taking a guess, based on this extremely limited dataset, as to which variables Quora was weighting for in determining relative ranking in an question set, I would guess the first would be total number of answers. People with more answers are more likely to have their answers ranked higher. I’d also speculate that the number of followers a person and the number of people a person followers has plays a role in ranking.

Quora does not track gender data of users, but in case there are some underlying community culture things going on, I like to track this as a variable. (I’ve seen speculation that women are more likely to be subjected to downvoting.) There were too few occurrences where differences occurred to speculate if something was going on. The two occurrences where there was a gender difference between positioned answers, the male answer with fewer upvotes appeared above the female answer.

Other statistics. The correlation between position and the number of votes is -0.4355, which suggests there is a slight relationship where the more upvotes you have, the lower ranked an answer is likely to be. The correlation between views and position is -0.466, which again suggests a slight relationship.

The correlation between the number of followers and the position is -0.268. The correlation between the number following and the position is -0.140. The correlation between the number of answers and the position is -0.274. The correlation between the total words in an answer and position is -0.052. This is pretty random, which suggests pretty much no relationship between these variables and positioning of the answer, despite what the totals for matched pairs say.

At this point, I’m honestly not convinced that that downvoting plays a huge role in relative positioning of answers. There are variables I suspect play an important role, and if there was a larger dataset, it might become much more obvious. I do think the socialness of people answering questions likely is weighted by Quora, potentially more so than downvotes.

Written 25 Dec.

Writing longer content for upvotes

May 17, 2015 Laura Hale content length, quora

Originally written for Laura Hale’s answer to Is it true that some people write long and detailed answers just to get upvotes and attract attention?

The real question though is how successful this is as a strategy for actually procurring upvotes? What is the correlation between upvotes and word count?

I went and found 130 different answers, recorded their word count using Word Count Tool – Free Tool to Count Number of Words. Word counter! and the total number of upvotes. I avoided new questions, to make sure questions had time to get responses. The correlation is 0.697, which is a pretty strong correlation. That number says: Yep, you betcha. In general, you’re going to get more upvotes if you write longer answers.

The average answer has 324 words. This is about three short paragraphs. To give an idea of that length, Kate Lang’s answer to What does it feel like to have OCD? is 311 words, and Biju Ashokan’s answer to What is wrong with Manchester United? is 328 words long. That number is really high though because of some of the outliers, including one answer of 5,400 words. The median answer size is 128 words, which seems much closer to reality in terms of actual response length. What does that look like? Lucas Braun’s answer to What are some cultural faux pas in Spain? is 133 words long. Ward Chanley’s answer to Does a therapist hold a right to question my non-theism? is 124 words long. That’s three shortish paragraphs.

The first graph uses logarithmic axes, because you cannot see the individual data points so well. The second graph does not.

When we put this data on a graph, you begin to see the misleading nature of the correlation. It is those pesky outliers to a degree causing that correlation. (When entering the data, one number took the correlation from 0.3345 to 0.8991. That’s the outlier influence. It is also why I reject average for this analysis, because they unrealistically skew.)

The better view might be quartiles. Let’s go with that. The shortest 25% of answers have a median length of 21 words. That’s pretty darned short. The next 25% have a length of 62 words, substantially longer. They both have a median upvote of 9. That little bit extra is not enough to get you more upvotes. The next quartile has a length of 225, which is substantially longer again. The value upvote wise based on the median? 11. The longest 25% have a median of 519 words, and gets 16.5 upvotes. For the normal writer, that is a difference of a grand total of 7 upvotes.

That is where you begin to see the futility of writing more for upvotes. They payoff is not there as evidenced in the table above.

The thing with something like this is Quora is a series of different microsystems at play. The community around questions about Spain is different than the community around Indian women is different than the community around Astronauts. Upvoting behavior can change from group to group. Who answers changes from question to question.

Let’s compare male versus female responses to see if one gender has an advantage over the other when it comes to getting upvotes for longer answers. There were 36 women who had responses recorded and 94 men. Yeah, Quora has a gendergap: There are not equal male to female response rates on Quora.

In any case, the correlation between a woman’s wordcount for her answer and upvote numbers is 0.009. That is really, really, really random. Increased word count does not help women. For males, the correlation is 0.71, which suggests men writing longer answers helps in garnering upvotes.

Strangely, it looks like women are also writing longer answers then men. Their median answer length is 163.5 words to men’s 121 words. They are also getting more upvotes on the median: 12.5 to 11.

At this point, quartiles are going to matter.

This is one where you go, “The data needs more women to double check this.” Women appear to be rewarded more for upvotes with shorter answers. Their answers get longer and they start paying a penalty, with fewer upvotes. (Note to self: Write shorter answers.)

The situation for men is different. They are appear much more likely than women to get awarded for long answers with additional upvotes, a signifigant number upvotes and way more than their female counters across all categories.

So conclusion: Writing longer answers may get you more upvotes but it isn’t always a sure thing. There are clearly variables like gender that may impact your ability to get more upvotes because you wrote a longer answers. On the whole though, writing longer for the express purpose of getting longer answers is probably not worth the ROI.

Written 21 Nov

Women and feminism: Male versus female answers

May 17, 2015 Laura Hale feminism, gender gap, quora

Normally, I do a lot of research about Wikipedia as a hobby, but I’ve largely been playing on Quora for the past six months. Like Wikipedia, Quora has some obvious problems with a gender gap and misogyny. There are a number of questions that also demonstrate that female users face unique problems that male users do not face: Creepy users of the opposite gender sending them messages based purely on their user name and their attractiveness found in their user picture.

There is a body of research that shows clear differences in gender usage of social media networks. A lot of this research actually goes a long way towards explaining some of the gender gap related issues for Wikipedia and Quora. A quote from a Forbes article, What Men And Women Are Doing On Facebook, sums up the gender use difference best:

However, women don’t just visit different sites from men, they use social media differently than men. Experts believe the difference between how men and women operate online mirror their motivations offline. While women often use online social networking tools to make connections and share items from their personal lives, men use them as means to gather information and increase their status.

…

According to Bowen, these gender differences are rooted in communication styles learned since birth. “Girls and boys are often raised in two distinct cultures where they learn different rules and norms for behavior and talk: Girls learn to build relationships by sharing social information. Boys learn to compare and compete with others, always striving for more success.”

What does this mean in terms of Quora, specifically as it pertains to the very female centric topics of women and feminism? I wanted to find out.

I’m not a programmer, so I couldn’t pull data so I had to resort to manual data gathering. I got responses 422 responses to 57 different questions. 22 of these questions were in the feminism topic, 23 were in the women topic, and 12 were in both topics. The topic, question, total number of answers, answer rank, user name, user upvotes, gender, and user followers were recorded. There are a lot more potential variables but given the desire to have a larger dataset, more variables were removed. Question and answer sets were collected until such a time that correlation for male and female for followers/upvotes did not fluctuate more than 0.05 after adding 10 answers. In case of anonymous users, their answers were not included as gender could be determined. Collapsed answers were recorded with the answer ranking showing up as higher than the total number of answers. Gender was determined first if it was stated in the profile, then by name (using baby name websites to assist) while looking at the picture, and by seeing if the answer included the person’s gender. In the 422 responses, there were 205 unique male users and 126 unique female users. The max number of responses for any single question was 37 as there was a desire to provide for a diversity of question, as there a lot more questions with fewer answers than there are questions with huge numbers of answers, which are actually outliers in this case.

So what were the results?

Operating Quora while female on feminism and women’s topics does not appear to cause any distinct advantage or disadvantage compared to male peers in terms of upvotes. Both are somewhat random, with what could be called a general trend of both genders getting more upvotes by virtue of having more followers at the same rate, 0.43 correlation for women compared to 0.41 for men. The slope of the line for women is actually steeper.

What does this mean? It means that while having a large number of followers may get you a “large” number of upvotes, this is not always the case. As you can see, there are a number of answers for people with 10,000+ followers who did not even get 20 upvotes for their responses. That rich get richer premise on Quora? Yeah. It doesn’t always play, and it doesn’t play for both genders.

On the other hand, when looking at the average number of upvotes by gender for males and females with more than 1,000 followers, females average 70 upvotes per answer compared to 37 for males.

Where things get a bit interesting is when we break down gender responses for women, feminism, and women and feminism related questions. Questions on feminism appear to have more randomness built into them, and this trend is the same for both females and males: 0.36 and 0.31. On women’s topics, the correlation is 0.51 for females and 0.47 for men. Women are rewarded slightly more for their answers about women than men are. When questions that are in both categories are counted, their correlation is 0.48 for females and 0.44 for men. Still a boost.

If we look at only answers by people with 1,000 plus followers, another gender pattern appears. Males with 1,000+ followers average 36.9 upvotes on questions about feminism. Females in comparison average only 14. When females with 1,000+ followers answer questions about women, they are at a huge advantage over male answers, 76 upvotes on average compared to 46.2 upvotes.

If we separate out only the questions that mention India, there is a significant difference in the results on this question. For all Indian related questions, there is a correlation of 0.63 between followers and upvotes. For females, there is a correlation of 0.69 and for males there is a correlation 0f 0.62. The more followers someone has, the more responses they are almost certainly likely to get. Here, the rich get richer seems to very much be a truth but the numbers are so small, that the relative number of upvotes almost is insignificant.

Upvotes aren’t all that matter. Placement matters a great deal. Which gender has the number one answers? Males edge out females in questions with more than one answer, with 34 number one placings compared to 19 for females, which is about 2 top male answers for every top female answer on questions related to feminism or women.

For topics labeled both feminism, women, there is no difference. There were 5 top answers for males, and 5 top answers for females. On feminism topics, males had 68.4% of the top answers, or 13 top answers compared to 6 for females. On the women’s topic, males performed slightly worse by having only 66.6% of the top answers with 16 to females 8.

Even if females are attracting upvotes on these topics because of their follower counts (or not getting upvotes because of their lack of followers), their answers are not being found at the top. For both men and women, the average number of answers where they had the top answer for a question was between 4 and 5 responses. The number did not change significantly when mean number of responses was looked at.

There is clearly something going on with the male to female top answer ratio. Males answers made up 61.1% (258) of all answers, while females made up 38.8% (164). Males are rising to the top about 5% more than they should as representative of the population.

Perhaps one of the reasons that women do not have top answers is the questions that attract males to answer are different than questions that attract women.

With 51 questions with more than one answer, ten only had responses from men, while no questions were only answer by women despite the fact that questions asked of women are included in the sample. If these ten questions are taken out, the ratio of male to female top answers changes to 24 : 19, putting females as over represented as proportional their population at 44% at top compared to a population of 42% of answers once the total male answers to questions only by males are removed.

The questions that attracted only males to answer were split with 5 for feminism, 3 for women, and 2 in both topics. The questions at the time of data gathering included:

One of these questions was subsequently deleted. One is aimed specifically at men. One is about finance. One is about sports. Several questions implicitly condemn women or their actions. One asks about hotness of women.

Which questions were more likely to attract women? Seven questions had 50% or more of their answers from women, and all of these questions had top answers written by women. Four were in the women topic, These questions included:

Three of these questions contain implicit criticisms of both genders with women as the cause, including the first, second and fourth. One question directly criticizes women. One is about dating. One is about boys. None of these questions are about hotness, finance or sports.

While questions pretty much implicitly asks women to talk about their personal experiences, males answered the questions. (For reasons why, see If you are a man, why do you answer questions that specify “if you are a woman” in the question?)

The more answers a question has, the less likely it appears that women will provide more than 50% of the answers in women and feminism topics. The threshold here appears to be around 16, though this could very well be a function of the sample size not including questions with large numbers of answers. After 16, there was not a single answer contained more than 50% of answers written by females.

Is one gender more likely to get downvoted or collapsed more than the other? Is there a downvote/collapse brigade against one gender or another? Not really, no. There were 31 collapsed non-anonymous answers. 11 (35.6%) of these were female answers and 20 (64.5%) of them were male. This is within 3% of total response percentage for both genders.

This picture changes when individual names are looked at. 7 of the 11 total female’s collapsed answers were from one individual, while only one male had an answer collapsed more than once with 2 collapsed answers. This does suggest that male answers are more likely to be generally downvoted while women are less likely to be collapsed on this issue. At the same time, the data suggests women are more likely targeted by so called “downvote brigades”.

This data paints a picture that shows there are clear differences in male and female answerings questions, and audience responses to answers by male and female question writers. Males are more likely to answer questions in both areas than females, and may be rewarded with top answers more than females just by sheer volume of their participation. On feminism, male are more likely to be rewarded with upvotes and top answers than their female counterparts. While female answerers may be advantaged with upvotes on questions about women, males get more top answers on this topic. Downvoting and collapsing answers may more broadly impact men, it specifically impacts individual women at a higher rate, suggesting that downvoting cabals exist that target women in this area.

Written 2 Oct, 2014.

quorastats

Some statistical analysis of Quora. All these posts are written for fun, because "mmmmm. tasty data!"

Alternative political views get downvoted (Puh-lease!) because they are unpopular views: An analysis of downvote collapses related to Charlie Hebdo

Introduction

Methodology

Results

Conclusions

Footnotes

The ’15 Top Writer upvote benefit: It is not as big as you think

Introduction

Methodology

Results

Conclusion

Footnotes

Appendix

The Great Quora Downvote Cabal does not exist: Characteristics of collapsed answers

Introduction

Methodology

Results

Conclusion

Footnotes

Appendix

Upvotes: Your answer is not going to get the vote total you are looking for

Introduction

Methodology

Results

Conclusions

Appendix

’15 Top Writers: Topic Area Expertise: Who knows and what do they know?

Introduction

Methodology

Results

Conclusions

Footnotes

Understanding the ’15 Top Writer community: Network analysis

Being a top writer means I am in an awesome upvoting clique! Right? I mean, really?

Variables for predicting answer placement: How much does the downvote likely matter?

Writing longer content for upvotes

Women and feminism: Male versus female answers