We've got a new statistic on the shiny new Profile Page Prototype. It's not perfect yet. And we did a pretty crap job of explaining what it's all about! Let's talk.

What you do here isn't just about solving one person's problem.

A while back, a user contacted us about his friend's account. His friend had been the top user on one of our sites, and active on a couple of others, and had died tragically and unexpectedly. We pointed him toward our guidance for these situations, but we wanted to give the family some perspective on how much their loved one had given here. It struck us that, unlike someone who donates their time in more traditional way (like tutoring), his family probably had no idea what a positive impact the user was having by helping so many people looking for answers.

I was easily able to pull up that he'd answered over 400 questions in just two years, and that the majority of his answers were deemed to be the most useful one. I could try to explain that he'd earned almost twenty thousand rep in that time. And I could tell them that what that meant was that he'd gotten roughly two thousand "up-votes." The problem was that all those things failed to convey what felt important:

A huge number of people who needed help very likely got it, thanks to this user. Digging in further, I discovered that his answers were viewed more than 590,000 times, with the vast majority of those views from people googling for help. That's the impact he made with his time here.

And the same is true for a good many of you reading this: the answers you've written and the questions you've shared have gone on to benefit hundreds, if not thousands of people who found themselves facing similar problems. This is a fundamental part of why Stack Exchange exists - and yet, nothing in the system recognizes this or communicates the multiplicative impact that individuals like you make here.

You gotta get this to get us.

Why should we care if people understand just how important the idea of reusable content is here? For existing users, it can provide some rough perspective on how big an impact your time donated here really has. One thing that drew me in long before I worked here was the idea that my effort here would help more people than it ever could buried in some forum. Now, it's fine if you personally find it just as fulfilling to help one person as it is to help a thousand. Or perhaps your motivation is the challenge in solving an especially hard problem... That's awesome too.

But whether or not the number of people who benefit from a post speaks to you personally, it's valuable for everyone to understand that the engine is designed to make content as reusable as possible, because it informs so much of how our system works...

See, once you realize that it's not just about the one asker, or the four voters, but rather the fifteen thousand searchers with the same problem... Well, suddenly, a whole ton of things that seem prickly and self-important, like editing out "Thanks in advance!" aren't about being "a bunch of power-happy pedants," they're about helping all the people who will ever have that problem to find the best answers. And find 'em instant-like.

How it currently works (and why).

Our initial idea was to capture "Approximate Views of your Helpful Posts". The concept was to give you some sense of how many people saw your work, and had a decent chance of having found it useful. It was not intended to be a replacement for reputation, or to grant any additional privileges within the system (as rep does). It's also not being shown anywhere but the individual's profile (unlike, say, badges). It's also not about measuring effort - we're working on some other ways to better display all the indispensable things folks do to improve the site, and will share them for feedback asap. This was to show the reach your effort had.

The initial methodology idea was just to sum up the total views on all your posts. But we ran into two key issues on answers:

  • We didn't really want it to include answers that no one in the community had deemed useful.
  • Plus, we can't really even count the views on answers, nor can we count just the views on a question page that came in after a given answer was posted.

So, we realized we had to start by defining some key goals and constraints:

  • Will be an estimate, but will generally be a solid approximation in most cases that minimizes meaningful anomalies
  • Can be applied to existing posts with data we have (meaning we don't need the post-answer viewcount)
  • Should limit answers counted to those that someone in the community found helpful
  • Should minimize cases where an answer that adds little (and probably isn't seen by anyone) adds large numbers to the count
  • When still in doubt, errs on the side of including views - we'd rather slightly over-appreciate volunteers then under-appreciate 'em

Ultimately we reduced our criteria to:

Views of pages where your helpful posts had some decent probability of being seen.

That doesn't mean we know the visitor scrolled to your post, or that your answer helped them - we simply don't track that. But we should be able to estimate the likelihood of a given question or answer been seen as useful by viewers without enabling the 25th answer on an insanely popular post to get a lucky up-vote and credit all the views to the new author.

That's how we got to the current method, which counts views on the following:

  • Questions
    • Non-deleted only
  • Answers - Views of the parent question for answers that are:
    • Non-deleted AND
    • Score > 0 AND
    • Also meets one or more of the following criteria:
      • In the top 3 answers OR
      • Is the Accepted Answer OR
      • Score at least 5 OR
      • Has at least 20% of the total vote count

(Note on closed questions: we originally were excluding closed questions, too, but it seemed to cause more harm than good: the ones that never should have been here almost all get closed before attracting many views, so including them very rarely matters. But the ones that are closed with a ton of views almost all were permissible back in the day, so having the score plummet when the rules evolve seemed more harm than the alternative, since the impact is so small there.)

This query is a very close approximation of what's live in the prototype. (It has some minor tweaks to make it play nicely with the public data.)

How can we make it better?

We need your help. Let's use this new number to help convey the underlying concept behind most of what makes us seem so crazy to new users, and to remind the rest of us why we (hopefully) think our time is better donated here than on any of the many other places we could be sharing what we know online. But let's do it right. That's where you come in.

Whatever we change, we're going to add a tooltip that clearly conveys that the number is a rough estimate, and also explains how it's calculated (or links to that methodology).

A lot of you really liked the idea of this number, but it elicited two specific types of helpful feedback:

  1. The methodology definitely needs to be clearer, and might be improvable
  2. Calling it "Bunnies and Puppies Saved by Hugs" "People Helped" is, at a minimum, a little distracting, and at worst, makes us "Filthy, evil liars." Okay, no one quite said that. But almost.

So, we want your input.

Big Question #1: Do you have any suggested tweaks to improve the calculation method?

Again, what we're going for is something that will generally approximate: "Views of pages where your helpful posts had some decent probability of being seen."

Remember that:

  • This isn't pure/perfect/science/a-number-that-actually-does-anything.
  • It is an estimate.
  • It's okay that it can't separate views that came before your answer, so long as it's very unusual that an answer posted way after the question on a popular question will count.
  • It's okay if it can get weird occasionally, as long as those cases are fairly rare, or will tend to occur when the number won't change that much anyway.

We think the current calc does a solid job at this, and most of the feedback so far was more about the name not matching the method (vs. the method being off,) but I bet you can help us make this estimate at least a bit more accurate.

Big Question #2: What's a better name?

We're aiming at something that:

  • Conveys the real-world, human difference you're making - "Page Views" doesn't do this well. "People helped" clearly evokes that concept, but the name...
  • Shouldn't include too many iffy assumptions about how people feel - If it suggests we know a lot more than seems possible, it clearly distracts from the goal. ("4,052 Babies Rescued From Killer Bees - All Thanks to YOU!")

The goal is to find a name that helps convey that the number is a ballpark representation of the people who had decent odds of getting some real use out of the effort you put in here.

Let us know what you think on the questions, and we can get to work improving this thing.

  • 64
    Sanity Check: So I post the 73th answer to a question with a million views. A year later it gets to 5 votes. Does that mean I helped 1 million people? – Mysticial Dec 2 '14 at 18:57
  • 6
    @Mysticial, no! That's why we're asking for help with the name, at a minimum. – Jaydles Dec 2 '14 at 19:06
  • 22
    What ever happened to the anonymous survey data? The "was this helpful to you?" question shown to users not logged in? Is that still being used? Could that data be used to reinforce this stat? – Won't Dec 2 '14 at 19:19
  • 4
    Can you provide us with a Data.SE query that lets you calculate the number on any site? – Martijn Pieters Dec 2 '14 at 19:25
  • 5
    I tried the query. "Timeout expired. The timeout period elapsed prior to completion of the operation or the server is not responding." – Travis J Dec 2 '14 at 20:00
  • 10
    I'm not really sure that we even need this stat. Right now it seems like a number to say "Wow, you're helping a lot of people!" But really, it's only for that individual. Why should any random person from Google care how many people I've helped? Why should anyone? If I'm helping a lot of people, many people probably already know my username and don't need a number to tell them that. And we already have reputation to tell someone if they're generating useful content. – hichris123 Dec 4 '14 at 1:33
  • 10
    SE isn't for helping people and if it is going to be, I'm leaving. Remove it. – bjb568 Dec 4 '14 at 1:39
  • 26
    @bjb568 so what SE is for? Insulting people only? Just wondering. – Shadow Wizard Dec 4 '14 at 23:55
  • 9
    @ShadowWizard Stack Exchange is for building a library of detailed answers to questions about specific community topics. – bjb568 Dec 5 '14 at 0:22
  • 6
    I think your paragraphs entitled "You gotta get this to get us" truly encapsulates what SE is for me - I know that I will be quoting from it for some time to come - brilliant writing! – PolyGeo Dec 5 '14 at 0:30
  • 10
    There's a little of the gamer in me thinking, "Well, if I find the top viewed questions with 2 or fewer answers, and write a good post for each..." Easily fixed, though, if you include post vs answer age. It's a simple calculation and you probably only need to trigger it for posts with more than 10k views. AnswerViews = PostViews * (AnswerAge/PostAge). – Adam Davis Dec 5 '14 at 1:02
  • 26
    @bjb568 And what is the purpose of building that library of detailed answers if not to help people? – Cascabel Dec 5 '14 at 7:49
  • 8
    @Jefromi The library is to help people, of course. My point is that focusing on helping people won't make as good as a site as trying to create a library of helpful information. Creating a library of helpful information focuses on quality. – bjb568 Dec 5 '14 at 12:14
  • 14
    I have a fundamental problem with this. At least since the "Hot Question" sidebar we all know that it's not the good questions and answers which get views, but those with catchy titles and/or shallow content. I'd rather see consistent quality rewarded. (Whichever way you'd measure that.) In any case, a "you were helpful" stat only based on views -- ignoring votes and anonymous feedback for the most part -- does not make a lot of sense. – Raphael Dec 5 '14 at 14:47
  • 10
    I find it curious that this seems to include page views that resulted in negative anonymous feedback (or even downvotes). Pop in, read the question, find it confusing, click on the down arrow, leave without reading anything else - and bingo, you've been helped by three answerers! – E.P. Dec 5 '14 at 21:41

32 Answers 32

The phrase "people helped" doesn't concisely convey what you described above because it appears without context. If you change the order of the stats, it may become a bit less ambiguous. Try switching the order to: "60 questions, 195 answers, 207k people helped." That's a bit clearer.

But this is about "reach". You should say their contributions have reached 207,000 people.

If you add a bit more context to the stats (below), it will also be a bit easier to understand. The presentation should look something like this:

Profile: people helped

I really like that last section. It drives home the ideas that these posts are reaching people outside the membership of this site.

If the example above is too wordy (or if you prefer the "helped {x} people" motif), I have a few other versions I was playing with. Maybe you can mix and match:

These posts have helped

~207,000

people searching this site

Or something that works when viewed in the context of the screencap above:

Has helped

~207,000

people through this site.

Hopefully between them you have something that better conveys the sentiment you are driving towards.

  • 22
    keep "reached", drop "helped", and then I start to agree that the stat is not a complete lie. Still, I'm not sure what the stat is good for. If we want to emphasize that the content reaches non-users, why don't we somehow visualize the anonymous feedback? – yo' Dec 2 '14 at 21:16
  • 44
    This is really smart thinking – what if we simplified it even more and just said "N people reached"? I think reach has an implication of helpfulness but is less explicit and so allows for the fact that some of the views might be from people who didn't actually find what they're looking for? – Laura Dec 2 '14 at 21:16
  • 6
    @tohecz Anonymous feedback still requires people to choose to interact with the page. Anecdotally, I know that a lot of people gain value from reading questions and answers on SO and never bother to vote or give anonymous feedback – that's what we're trying to capture here. – Laura Dec 2 '14 at 21:17
  • 1
    @Laura Which just means that we can think: Ok, only c. one in ten people use the AF. Then you can really measure the outreach. But without that, the number is just meaningless. – yo' Dec 2 '14 at 21:21
  • 3
    @Laura Just 'people reached' will not be understood by a lot of users. Especially users outside of SE. – David Mulder Dec 3 '14 at 7:10
  • I was going to suggest something like "exposure level/rate", but this suggestion is far better. Cheers and hope to see it implemented. – Shadow Wizard Dec 4 '14 at 23:58
  • 1
    @Laura The whole problem is that until very recently, nobody has ever mentioned what is the stat supposed to be good for. Now it's clear that nobody cares about it's preciseness or whatever, it's just an advertisement, and we live in a society where adverts lie. Fine. – yo' Dec 5 '14 at 0:29
  • 14
    +1 for "These posts have reached ~X people searching for help", tilde included to indicate it's an approximation. It works and doesn't make much up. Please don't call it "These posts have helped ...", because that's back to presuming views equate to help provided, which is the lie so many of us have been uncomfortable with. – doppelgreener Dec 5 '14 at 1:23
  • 2
    I would prefer 'searching for help' to 'searching this site', as I suspect the majority of people landing here were actually 'searching Google' :) – Benjol Dec 5 '14 at 8:48
  • 1
    I'd prefer \approx over \sim. – Raphael Dec 5 '14 at 14:44
  • @Raphael That would imply more certainty about the metric than is probably warranted. – Air Dec 5 '14 at 17:37
  • @AirThomas That may be "cultural"; in my field, \sim is stronger than \approx. – Raphael Dec 5 '14 at 18:26
  • +1 I like it. But one can worry less about the precise wording if there's a link you can click to understand the metric and how it's calculated. – HostileFork Dec 5 '14 at 23:09
  • Correlating question age, question upvotes, question pageviews and (total/average/(rms? I dunno)) answer upvotes sitewide might be useful setting the right vote count. – jthill Dec 6 '14 at 0:15
  • At first, was thinking "People Educated", but they may not have learned from the post. But why not say "People Coached", because that is what we are trying to do. – Sablefoste Dec 6 '14 at 5:00

Name: Impact or Reach.

On the algorithm, I'd like to see some use of specific-post links. When people follow a link to this specific answer, it feels like that should mean something even if it happens to be answer #4 and wouldn't count otherwise. It doesn't make sense to just add the view count from those links (it's usually a very small number), but can we find some way to factor it in? Maybe one of these as an additional option in the "one or more of these criteria" list:

  • Has had at least some threshold (25, per Announcer?) visits to this specific post (may be expensive to compute)

  • Has had an Announcer badge awarded for it (to anybody) (easy to check but may set the bar too high)

  • 2
    I like both the names. The link criterium makes sense, but may not make much real difference - the views coming from direct links to answers are really few, and I'd bet the ones on answers that wouldn't already meet the criteria would be too small to matter much. – Jaydles Dec 2 '14 at 22:08
  • @Jaydles true, I conflated two stats there. Counting the referrals doesn't add a lot, most of the time. If a post is getting referrals, though, then maybe it should get some extra love even if it's not in the top three (etc), so long as it's still positively scored. Needs more thought. – Monica Cellio Dec 2 '14 at 22:14
  • 1
    @Jaydles I made an edit; does that help? – Monica Cellio Dec 2 '14 at 22:27
  • 7
    Yes, impact is the term already used for this in traditional media (academic journals and so forth) – Ben Voigt Dec 5 '14 at 2:53
  • The word "reach" was mentioned several times in the question. I think "reach" makes as much sense as anything else. – Bryan Oakley Dec 5 '14 at 22:44
  • @BenVoigt impact is used in academic literature for getting your paper cited. So, there is some proof that it did literally have impact - it was not only "given" to the other person, but the other person "took" it and incorporated it in his work. "Reach" assumes much less - it only says that the message arrived at the receiver. It can be that the receiver disregarded it. "Impact" for me assumes that it had an effect after being received. – rumtscho Dec 29 '14 at 19:03
  • @rumtscho: Unless you're talking about getting published in a journal with high impact factor, in which case it is all about the number of eyes that will see the new article. – Ben Voigt Dec 29 '14 at 19:05
  • @BenVoigt an "impact factor" of 3.14 means that the articles in this journal get cited 3.14 times on average. While it does correlate with eyeballs, the primary information, and the formula used to measure it, is based on the number of citations, and to me it sounds natural that getting cited is a proof for having impact, while landing in somebody's mailbox (in the case of a journal) isn't, it's just a matter of having reached somebody. This is why I find the "reach" wording good and the "impact" wording misleading. After all, the journals don't calculate IF using number of abonnents. – rumtscho Dec 29 '14 at 19:23

The folks at Academia SE already know the name (although the concept is different there):

(drumroll)

Impact factor

The formula should IMHO be modified in a very simple way: take a decimal logarithm and multiply by 10 (the familiar deciBel paradigm).

Why this log-transform? To make people forget about maximizing/tracking this number, and to let it improve visually only after an order of magnitude improvement in impact.

NOTE:

The hidden variable SE devs are trying to measure is severely underidentified - I see no mathematical/statistical possibility of deducing an answer's impact from question views. Even with JS "reading time" statistics, which may be a bit too intrusive to my liking, the error margin is just too wide.


As I have said before there's little value from making the number discussed a direct competitor to reputation. Here's another take on the profile redesign.

  • 23
    If they do that, I'm gonna change my username to "meteorite that wiped out the dinosaurs" and ask why my impact factor isn't higher. :D – Scimonster Dec 2 '14 at 21:47
  • 5
  • 3
    Taking the log is a good idea. It makes it simple for humans to compare things that differ by orders of magnitude. – Edward Dec 5 '14 at 0:13
  • 8
    -1: I doubt many users will be comparing their impact factors with other users'. They will mostly be looking at their own number. And making that number harder to interpret by requiring users to understand a logarithmic, non-obvious formula will make the number less useful for individual users. – Kevin Dec 5 '14 at 2:18
  • 9
    @Kevin making "the number less useful for individual users" is precisely what I like about this proposal. We've got rep whores already, do we really want impact whores? – gnat Dec 5 '14 at 8:13
  • 1
    @gnat If we did, they'd have a really hard time improving their impact factor after a while, since it increases logarithmically. – doppelgreener Dec 5 '14 at 8:54
  • 3
    @doppelgreener right, logarithmic increase is what I like here, just for the reason you mentioned – gnat Dec 5 '14 at 8:57
  • 1
    I cringe when I see "impact factor" because this term is heavily abused in the academic world. "Impact" is somewhat vague and cannot be measured precisely. Consequently precise numbers give a false impression of precision. If you're going to give a precise number, best to measure something plainly measurable and give it a name that accurately reflects what it really measures, like "page views" or "reach". – Ben Kovitz Dec 5 '14 at 14:18
  • 4
    @BenKovitz - the formula set by SE devs doesn't reflect either page views or reach, and is IMHO superfluous. My post is an attempt to get some lemonade from a lemon. – Deer Hunter Dec 5 '14 at 14:40
  • 1
    @Dilaton Ah, I didn't know that. I've never been on the Physics SE. I've mostly been blown away by the wonderfully high quality on the original stackoverflow.com. Things have gotten to the point where I don't think I could program without it. – Ben Kovitz Dec 6 '14 at 0:31
  • 2
    I suspect that the new statistical measure will further enhance the impact of popularity with a general lay audience at the expense of usefulness to serious students, professionals, and experts which is naturally a smaller group of people. – Dilaton Dec 6 '14 at 0:33
  • 2
    @Dilaton - too true as far as supercollider questions go. – Deer Hunter Dec 6 '14 at 0:35
  • 1
    @DeerHunter I always wondered why the (infamous: Shog9 explicitely admitted that its intent is to entertain the network wide crowd which has nothing to do with high-quality) hotlist is called "supercollider", as the term supercollider rather makes me think of the LHC or the unfortunately killed by stupid American politicians SSC ... ;-) – Dilaton Dec 6 '14 at 13:24
  • 2
    @Dilaton - meta.stackexchange.com/a/66838/204631 – Deer Hunter Dec 6 '14 at 13:31
  • 1
    While as a scientist, I have no problems with logarithms, a central point of this stat is to convey your impact in some way understandable by somebody who does not know anything about Stack Exchange in the first place and in particular also does not know the formula behind it. And on top of that, there are people who are not sufficiently familiar with logarithms to understand it, even if they knew the formula. If we use logarithms, we might as well stick with reputation. – Wrzlprmft Dec 22 '14 at 19:51

Metric concerns:

Use stricter validation for late answers on posts with more than 100,000 views. Perhaps if it is a late answer on a post with more than 100,000 views, then it needs to have multiple criteria (such as more than 20% and top 3?5?).

Basically stricter metrics for late answers on high view posts to reduce gaming this number.

Naming suggestions:

  • Audience Reached
  • Helpful Post Views
  • Notable Post Views
  • 9
    I'm a big fan of "helpful post views". It's a pretty succinct description of a pretty complicated number. – Laura Dec 2 '14 at 19:13
  • 2
    @Laura I still don't see why my clicks on Qs and As are instantly considered as helpful. – yo' Dec 2 '14 at 20:01
  • 2
    @tohecz Because the large large majority of users come to SE to find solutions to their problems rather than just browsing for their own learning like we probably do. – David Mulder Dec 3 '14 at 13:30
  • @tohecz - If you browse to a question, and view the top few answers, and do not down vote, then you have at the very least indicated that the content you viewed was not "not useful". Many views do not even have the chance to vote, and often users will make new accounts with the sole purpose of being able to upvote because they feel they need to give back in some way. – Travis J Dec 3 '14 at 16:03
  • Travis - I'm not so sure that is valid. Many folks who come from outside SE may find the answer that sorts their particular need and then not upvote. You can't assume a lack of vote means a downvote (or upvote). It is just a null. – Rory Alsop Dec 3 '14 at 23:06
  • @Rory - I was specifically responding to tohecz "why my clicks on Qs and As". The downvote tooltip reads "not useful", and by not downvoting, it is at least obvious that it was somewhat useful. It is not just null, because users here are encouraged to downvote useless content. In your example of outside users, please see the latter point I make about those users subscribing to the network in an attempt to vote in the future. – Travis J Dec 4 '14 at 4:18
  • It would help accuracy if views were tracked separately for answers, so that a late answer to a once-popular question wouldn't "inherit" all the views the question had already collected in the past. (In fact, that would help with other things too, like with the algorithm used to turn off rel=nofollow for links in reputable answers.) Of course, the initial view count for old answers would have to be guesstimated somehow, but then, that's what we're already doing anyway. – Ilmari Karonen Dec 5 '14 at 0:54

I realize this might be undesirable in terms of screen space or visual aesthetic, but I would seriously consider using verbiage instead of a number:

  • Hundreds of people reached
  • Thousands of people reached
  • Tens of thousands of people reached
  • Hundreds of thousands of people reached
  • Millions of people reached

And further, if the reach is less than "hundreds", don't bother showing it; if it's more than "millions", still just say "millions". (The number of tiers and the level of the top tier are subject to tweaking, but it should be a small, fixed number of very coarse tiers.)

I'm proposing this because it balances a number of themes already expressed in various answers and comments:

  1. The thing we're trying to measure is necessarily imprecise (I would argue extremely so), with a lot of guessing and fudging.
  2. We want a number that relates to some easily understandable units (so not something like the Richter scale or Dow Jones Industrial Average). This is at least in part to satisfy the use case of being able to tell loved ones that their friend positively affected such-and-such number of people.
  3. We would like people not to obsess over this measure or try to game it.

I like the previously offered suggestion of using a logarithmic scale; my proposal just takes it a bit further (and explicitly removes the option of transforming it into some kind of unitless "factor").

Why not go for something based on a well known SE meme?

Unicorns healed: 4024

This has the benefit of a pretty absurd name which immediately suggests that this metric should not be taken too seriously. It also doesn't make any unbased assumptions about "people helped". It's a fun name that would ensure that nobody gets too worked up about what it represents while still conveying the relevant information.

  • Or Orcs vanquished? (I thought about trolls vanquished, but that raises a whole 'nother meaning). – bib Dec 5 '14 at 16:29
  • 2
    I'm not aware of any SE memes about orcs. – Lightness Races in Orbit Dec 6 '14 at 18:01

I found out a bit late of this follow up, but I will be re-iterating what I already wrote in my other answer, though this time round I focused purely on the calculations of the number.

Proposal

  • Questions
    • Non-deleted only
  • Answers - Views of the parent question for answers that are:
    • Non-deleted AND
    • Score > 0 AND
    • Multiplied by the factor of users that on average scrolls to answers in this position AND
    • Answer is posted within a week of the question IF NOT
      • Multiplied by the difference in dates (incorrectly) assuming a linear distribution of views

What is this solving: Answer position

Right now, a question with three answers will always count all answer, yet a question like this will count 17 (!!!) answers equally, whilst the large large majority of those 243723 viewers of that last question will never ever scroll past the first answer. So what I am proposing is that Stack Exchange temporarily adds tracking code that tracks actual answer views based on scroll position (this is really easy to program and they might already have this data) and next this data would be applied globally as multiplication factors. In other words, you would get a list of global factors like [1, .4, .2, .1, .09, .06, .......] for example with the average number of views answers in a certain position get. Of course, it's still an approximation (because of the generalization), but at least it's a FAR better approximation.

What is this solving: Posting date

Especially assuming that the above is for whatever reason not implemented then right now users will try to game the system1 by searching for posts with the largest number of viewcounts with at most two answers. If they post an answer and get at least one upvote they now grab the full views of the post. Alternatively from the previous solution - or per this proposoal; in addition to - it is also possible to assume that views come in in a linear fashion (absolutely not true, but it keeps it nice and simple) and distribute views that way. Additionally answers within one week would be counted fully, for the simple reason that that keeps the calculation faar cheaper (as a lot of answers are given within that timeframe). The reason for this approach is that it's possible to do these calculations purely with the data that is available now and would make the calculated number far more precise.

1 Which I don't think a lot of users will be wasting time on, as the metric is only shown on a profile tab users themselves will not commonly look at.

  • A v. good idea for data collection. – Deer Hunter Dec 3 '14 at 11:40
  • And to whoever downvoted this, care to explain what issue you take with these proposed improvements? – David Mulder Dec 3 '14 at 13:26
  • Great, more downvotes without anybody explaining... such mature behaviour... – David Mulder Dec 4 '14 at 17:50
  • 9
    This is meta, downvote means someone disagree with what you suggest. – Shadow Wizard Dec 4 '14 at 20:06
  • 1
    @ShadowWizard So, groundless disagreement is suddenly the norm? Both on the main sites and on meta there still is this expectation to clarify downvotes. – David Mulder Dec 5 '14 at 5:31
  • 5
    I did not downvote but no, there is no such expectation and there never was. That's why downvoting is anonymous. I think this is a reasonable suggestion but that doesn't mean that disagreeing with it is "groundless" by definition. – terdon Dec 5 '14 at 8:13
  • 2
    @terdon Ehm, so, if there is no such expectation, why then are new users shown messages to explain downvotes in the comments? True, if you're are extremely pedantic then you could argue it's just etiquette rather than an expectation, but etiquette is nothing but a cultural expectation, so that doesn't fly either. – David Mulder Dec 5 '14 at 8:30
  • 3
    @David no, anyone can vote as he/she wants. That's how it works. If you don't want to take part, nobody is forcing you to. – Shadow Wizard Dec 5 '14 at 9:09
  • 1
    @ShadowWizard Of course nobody is forcing you to, expectations are just that, expectations. I dunno, one isn't forced to eat with knife and fork, that doesn't mean nobody expects one to eat politely. – David Mulder Dec 5 '14 at 9:13
  • True, but when someone is eating with an open mouth others are free to express their disagreement. – Shadow Wizard Dec 5 '14 at 11:15
  • Not sure I like the scroll position metric -- there's got to be too much variability on that one. Pro-rating by date posted makes sense. It will still probably tilt in favour of late answers, since many views are in the first week, but it seems to be the best we can do with the metrics that are already tracked. – AmeliaBR Dec 5 '14 at 15:51
  • 2
    @ShadowWizard Which is exactly what I did and what you attacked me about? – David Mulder Dec 5 '14 at 16:40
  • @AmeliaBR I know there is a lot of variability on that one, but at least it's fairer than the top 3 metric right? I mean, with a lot of work you could make it a lot fairer, but I tried to keep in mind that this was just a small number on a profile page, not some main feature of SE and all things considered I think this would strike a fair balance. – David Mulder Dec 5 '14 at 16:41
  • Another answer suggests to use time distribution of upvotes as proxy to calculate a view factor. – Nemo May 4 '15 at 8:59

As I posted before this thread was started (undeleted by now):

How about a simple "post views" with as little black magic as possible? The term uses words that are well established and understood. "posts" is universally used as umbrella term for questions and answers. (Let's keep ignoring comments for simplicity.) And "views" also has very little potential to be misunderstood. It corresponds to the term "viewed" we see for any question. And we already have "profile views".

Since you cannot accurately measure views for answers, there has to be some black magic. You could take total views and multiply it by a factor 0 < x <= 1. Doesn't have to be all-or-nothing. Could still be a cheap calculation.

Much of what you have now would go into that factor. But with the neutral term "post views" we could strip assumptions about whether a view actually might have helped the viewer. It's hard enough to pin down the number of "views", especially for answers, but even for questions (think deep links).
Still: less speculation. In this spirit, a single answer to a question with negative votes could still just count fully - adjusted for the point in time it was posted.

To account for the point in time when an answer was posted, we could figure out the typical waveform for total post views over time with statistical methods and then pick a simple formula to best approximate.

Or we could go out on a limb and just save the number of views at the time of posting to remove one source of error. Could be another int column PriorViewCount (like the existing ViewCount) in the table Posts. The great majority of posts only has up to three answers. You could then get mostly accurate numbers for those. For the rest, you'd still apply a formula for the approximate share of views.

Disregard deleted posts (even though there were views before they got deleted).

If this thing is popular enough and you have a spot to spare you could supply two numbers: "total post views" (according to the above) and "estimated helpful" (the number after applying more black magic).

P.S.: When I wrote "but it's really a lie", that wasn't meant to imply you were "filthy, evil liars". Your effort to make the profile more useful is well received and appreciated! But people are typically more than happy to believe flattering news. Not few would start boasting about the "millions" they supposedly have helped, which would quickly become ridiculous. There is no need for that. The impact of this site is great without exaggeration.

  • 4
    Lie as in "every stat is a lie". +1 from me. – Deer Hunter Dec 4 '14 at 22:46
  • Although post views is well understood within the SE community, it means quite little to an outsider... which per OP is exactly one of the main intentions of this metric. – David Mulder Dec 5 '14 at 5:29
  • @erwin, I totally know - the joke was meant to convey (playfully) that you had a point- thou were right that he perception of exaggeration undermined the actual impact. We appreciate the feedback, and I hope you know, the hyperbole was intended as a joke! – Jaydles Dec 5 '14 at 5:34
  • @Jaydles: Now, that must be another one of those sneaky, filthy lies of yours!!11 Nah, I think we're good. :) – Erwin Brandstetter Dec 5 '14 at 7:37
  • 2
    @DavidMulder: I can see the intention. I would have imagined the typical visitor of a profile page to understand the term. Either way, just a suggestion. Personally, I am more interested in a meaningful number. – Erwin Brandstetter Dec 5 '14 at 7:44
  • +1 This sounds like a great idea to me. If the stat is just the number of views a user has garnered, it's extremely obvious how the stat was calculated, and it's relatively easy for someone to understand the context by which that number should be understood and adjust it in h his or her mind. – Kevin Dec 5 '14 at 16:47

The problem is with "people"

The word, not actual people. As of today, the stat now reads

X people reached

I think there is still a problem, and it is not nor was not about the word helped or reached.

1 person can be 100 people

The reason "people" is problematic, especially when qualified with "reached" or even "helped" is that it implies that the people are unique. The metric should improve by 1 for each person I help.

This is not the case though, if one person has viewed 100 of my posts, suddenly they count as 100 people. Take into account that this can happen with more than 1 user, and the "people" that have been reached or helped is clearly exaggerated.

What or who was really helped?

The reality is that one person has had 100 quandaries, dilemmas, problems, or inquiries which led to them viewing 100 of my posts. I think that this is what should be related in the title, and that it still relates to people. People are the source of these quandaries, dilemmas, problems, or inquiries and it will still be a rough indication of how many people have been helped while not exaggerating the exact number of people.

  • 100 quandaries [ helped | assisted ]
  • 100 dilemmas [ helped | assisted ]
  • 100 inquiries [ helped | assisted ]
  • 100 [ helpeful | positive ]? collaborations

I believe one of these may be a more accurate description, while still conveying the helped terminology.

Definitions

Quandary: a state of perplexity or uncertainty over what to do in a difficult situation.
Dilemma: a situation in which a difficult choice has to be made between two or more alternatives, especially equally undesirable ones.
Inquiry: an act of asking for information.
Collaboration: work jointly on an activity, especially to produce or create something

How about keeping the algorithm simple:

                    (views on question) * (score of your answer)
"people helped" = ------------------------------------------------
                    (total votes on entire question and answers)

(I'm using "people helped" for now, but I'm not actually proposing that as a name.)

Other considerations:

  • "score of your answer" = upvotes - downvotes
  • "total votes on entire question" = upvotes + downvotes*

*Because all votes up or down came from a view. IOW, some views didn't actually help anyone (especially if they downvoted)


Pros:

  • Asymptotically meaningful.
  • More resistant to gaming on popular questions.
  • Encourages answering of unanswered questions with a lot of views.

Cons:

  • Potential for noisy "up-and-down" effects as explained in comments.

Possible Tweaks:

  • Replace "total votes on all posts" with "total upvotes on all posts". (suggested by Servy)
  • Include anonymous feedback.
  • Plenty of tweaks available: Such as including the anonymous feedback in the vote counts. – Mysticial Dec 2 '14 at 19:20
  • 6
    Using percent of answer vote as a multiplier will cause a bunch of weird behavior. The most common is that even if your answer is always on top, every new one that gets votes will reduce your number. – Jaydles Dec 2 '14 at 19:22
  • 1
    I like this a lot because it discourages people from going to the most popular questions and sticking in an answer. That becomes a rather poor strategy for upping this number in those cases. – Servy Dec 2 '14 at 19:22
  • 7
    @Jaydles Is that undesirable? I would think you should get more of this stat from a question in which you are the only answer than one in which half of the people felt that they needed to go to another answer to get help. Your answer was still helpful, just not as helpful. One case that does need to be dealt with though is that a new bad answer getting a ton of downvotes would reduce your perceived helpfulness. That doesn't make sense. Total votes might need to just be total upvotes. – Servy Dec 2 '14 at 19:24
  • Do we want this to add negative values for posts with a negative score? – Servy Dec 2 '14 at 19:25
  • 1
    @Servy, it just creates a weird experience - if you have a 4-voted answer, and then someone else gets a vote, your "Helpful views" or whatever goes down by 20%, which seems off somehow. And if we're that focused on potential abuse, this could drive a lot of weird voting incentives - adding the vote multiplier means that down voting every competing answer will move this number on almost every question. – Jaydles Dec 2 '14 at 19:27
  • 1
    @Servy That's a good question. If an answer is bad that it's wasting people's time, does that count has a "negative helpful view". My opinion is yes. – Mysticial Dec 2 '14 at 19:27
  • 1
    @Jaydles Using the formula in this post, downvoting competing answers would harm your own score (increasing total votes without increasing the score of your post or views). If it doesn't, then it would have no effect on your score. In either case it would indeed lower the score the answers you downvoted. – Servy Dec 2 '14 at 19:30
  • 1
    @Jaydles I think there's two (somewhat conflicting) goals here. A number that is meaningful asymptotically, or something meaningful at small scale. I propose this algorithm because is makes asymptotic sense. But it may cause weirdness at smaller scales. One way to solve this is to record the current "people helped" score for an answer each time it is updated. And never let it go below that. – Mysticial Dec 2 '14 at 19:31
  • 1
    @Servy, I think you're right - I was confused. But isn't that weird, too? If someone else adds a terrible answer, your score drops every time it's down voted? – Jaydles Dec 2 '14 at 19:34
  • @Jaydles That's why I suggested using total upvotes, not total votes. – Servy Dec 2 '14 at 19:55
  • This is asymptotically meaningful, but has nothing to do with helpfulness, alike the original concept. – yo' Dec 2 '14 at 20:04
  • @tohecz It has something to do with helpfulness, it's just a stat with a fairly low degree of confidence. That confidence is certainly non-zero though. – Servy Dec 2 '14 at 20:08
  • 1
    @tohecz No, but what it does do is distinguish the truly useful questions with a very narrow scope from the truly useful questions with a very wide scope. – Servy Dec 2 '14 at 20:28
  • 1
    Which brings us to the basic problem, which is that the metric measures popularity and not helpfulness. – yo' Dec 2 '14 at 20:33

I think this is a variant on what David Mulder suggested:

It seems clear that 'views' is way too high. I know that when I'm searching for help, I'll happily open the first 10 links in 10 tabs and often not even get as far as reading the last ones (or conversely, I can get to the end still not have been 'helped').

But do you have any statistics on how long people tend to stay on pages? And whether it depends on the number of answers (or just the length of text)? In that case you could apply some kind of global 'factor' to bring the view count down to something a bit more realistic.

  • 1
    I agree. A simple javascript setTimeout set to a reasonable duration for the content length before sending the http request to increment the count would be more accurate. There are many times that I find a post that looked like it was what I needed, but after a few seconds on the page I realized it wasn't and went back to google. Therefore the post wasn't helpful to me specifically at that moment. – SnareChops Dec 6 '14 at 7:05

I'd like to argue for restoring the people helped phrase while also making the calculation more conservative.

Let's acknowledge upfront that this number is a pleasant fiction—we can't poll everyone who visits every question to find out which (if any) answer helped the most. And there's no reasonable way to know if a post helped anyone other than by measuring votes. (We already have a system for estimating people helped from direct voting: reputation.) In almost every way it doesn't matter what number we pick as long as it goes up each time a user asks or answers a question helpfully. When numbers get large (over a few hundred) it's really difficult to visualize them. The only way we can relate to large numbers is to compare them to other large numbers.

The trouble with using the label "people helped" is that we have various measures of how many people visit the site. MSE is visited by fewer than 10k people a day, so it seems off to report a couple million people helped. Maybe that's a bad example; Jeff has probably helped more people than that in one way or another. But as you poke around at other people's profiles, you inevitably find folks who helped ~140k people with a Jon Skeet fact. Remember, this site only has 119k registered users.

So we can start eliminating edge cases like Community Wiki, which probably makes sense in any case. However, we are still going to have questions that get outsized views and skew the results in a way that breaks the illusion* that the number represents actual people being helped. Giving full credit to every view a question gets to every asker and many answerers breaks verisimilitude for me.

My suggestion is to scale down views using a scaling factor that reduces the impact of heavily viewed questions and divides credit between the asker and all the answerers:

log10(VeiwCount)^5

With this scaling, an answer to the Jon Skeet facts question potentially counts for something like 3600 people helped. Rather than use the proposed formula for who would get credit, I'd like to suggest that an asker gets credit for helping that many people their question:

  1. has at least one answer,
  2. has a score greater than 0, and
  3. is not Community Wiki.

An answer would get credit for helping people if it:

  1. has a score greater than 0, and
  2. is not Community Wiki.

Note that for often-answered questions (which are also often highly-viewed) this will credit more answers than the original formula. However, given the log scaling, the total "people helped" credit will be a lot lower. These criteria also prevents unanswered and negatively-scored questions from counting as helping the people who merely viewed the question.

Under this algorithm, Jeff Attwood will have still helped 379k people on MSE. Jon Skeet will have helped 9.6 million folks on SO (compared to 103 million). These numbers feel right to me. They are absolutely wrong, but they are also usefully plausible.


* The illusion is more important than any semblance of accuracy. The designer of the Civilization computer games, Sid Meier, notes:

Gameplay is a psychological experience: I base my games on things like railroads, pirates, and history, and I try to make the games I design true and real. The more historical, the more realistic, and the more factual, the better. The more railroady and piratey, the better. But what I thought I knew was wrong. During the early days of my career, I hadn’t taken into account what was in the player’s head. By acknowledging that simple concept—that gameplay is a psychological experience—it can make your games better.

  • I'm still reading this answer, so I can't honestly comment on the actual content....IMO, this post might be more visible as a new Q. ( older question, 26 answers, etc. ) – Shokhet Dec 18 '14 at 3:56
  • my reading of "people helped" name suggests rather precise measure; this seems to clash with heavily approximated nature of the calculation (FWIW approximation per se feels quite plausible to me) – gnat Dec 18 '14 at 8:57

Psophometric assistance factor

OK, OK. Don't look at me like that.

Here's the rationale:

  1. it's designed to be in line with human perception
  2. it's trying to show how people have been assisted
  3. it's an estimate
  4. it doesn't really have an obvious unit of measure

The word "psophometric" is fairly obscure, but it essentially means a measurement that's adjusted according to some model of human perception. For example, audio engineers use a "psophometric weighting" to quantify human perception of noise. We are searching for a similar kind of adjustment that matches our collective perception of this notion, and will adjust it until it (subjectively, but by some consensus) matches our concept of what it means to meaningfully participate here.

The word "assistance", while perhaps speculative (we don't really know that people were assisted, we infer it from indirect measurements) captures something about what we're at least attempting to quantify.

The word "factor" simultaneously conveys both the notion that it's not the whole thing but just one part, and the notion that it's unitless and not really convertible to furlongs or kilograms or cat-video-viewing-hours or any other physical measurement with real units.

The other thing is that because the phrase is very likely to be unique on the internet, anyone searching for the definition would likely get sent to the correct place(s) among the Stack Exchange constellation of sites.

  • I like the idea of conveying it as weighted, but I am not sure about your word choice and could not help but get stuck on your second sentence. – Travis J Dec 5 '14 at 0:37
  • 4
    It has a lot going for it, but doesn't it seem... Just a little too obvious? :p – Jaydles Dec 5 '14 at 0:38
  • I'm not sure of my word choice, either. Just throwing it out as a possibility to debate, modify, adopt or discard as seems most appropriate to the group. – Edward Dec 5 '14 at 0:38
  • +1 for learning me a new word. Also: PAF! :) – Benjol Dec 5 '14 at 8:56

Possibly off-topic though I believe it to be an underling factor:
Nothing other than vote count belies the usefulness of a post in the first place. Even then, one has to look at its age and compare it with views and votes. A five year old post with 100k views and +300 votes isn't any more pertinent than one from 6 months ago with 2k views and +20 votes. I have to perform this mysterious calculation in my head to determine a page's Usefulness Score.

For example this page's score would be through the roof at the moment, as it should be. Later, as it ages, it will become less so (useless to all but meta SE's posterity). Also, vote count means absolutely nothing. I'm somewhat of a stickler for how to vote, but even I still find myself voting for the wrong reasons; humor, haz pic, ect. Not always whether a post was "useful". +1 can mean all sorts of things.

You can't pull a new stat out of the air with the current data set. I'd say out of the ~1000 upvotes I've cast, ~5 are for something that was actually useful to me, as opposed to me judging that it might be useful to someone. We need a 'tick' like the favorites button for every answer; "this is the suggestion that worked for me". Who left it up to the OP to decide what is right? Where is the community's accepted answer? You know, the one under the accepted answer with 700 more upvotes than it... (limit one a day, or people will just click both when they upvote anyway) Perhaps some positive Flags would work too.

Here's a great example (although I don't agree with the content, numbers like this are what I mean): Gender-neutral alternative to “craftsmanship”? IMO, no one should ever find that page useful and I think the OP barely did themselves. The numbers on that page only tell us how uppity we can get. Nevermind what the hotlist can do to skew a page's numbers... that's a whole 'nother sack of beans.


TL;DR

You will get no "solid approximation" based off of meaningless anomalies (votes).

We have absolutely no way of knowing "that someone in the community found it helpful" (you delete those comments, remember?)

Any answer that adds little: ignore all "Late Answers" . Only people who don't care about rep late-answer.

Erring on the side of including views is meaningless; view doesn't mean I read it.

The closet thing we have is the favorites button. I use them as bookmarks when I find something I know I'll want reference to. Not to link later but for my own use in the real world. I think a lot of people use them to keep tabs on pages they're watching; still it's a better number to draw 'usefulness' from. However, you can only Favorite questions; not "this is my favorite answer, 'cause it worked for me". That information gets lost in the (phantom) votes.

Since an Accepted answer has a very high chance of actually being read (and even helping viewers), I think an Accepted answerer should always be credited -- even if the vote count was 0. We should not penalize the answerer just because the asker happened to be too new to SO to be able to upvote (or to know that they could).

So, I'd propose changing:

Score > 0 AND

to

(Score > 0 OR Is the Accepted Answer) AND

If this is done, then effectively we would credit the answerer of all undeleted accepted answers.

The only thing that would make sense to me is showing a statistic on the anonymous votes on one's posts. Reasons:

  1. This statistic is never shown and could be really interesting and significant.

  2. Registered users can vote on the Qs and As, so their opinion on the post is reflected in reputation.

  3. Non-registered users can vote through the anonymous feedback. Their visit itself may be very well irrelevant noise, and very often it probably is.

  4. I still don't see how views correlate to usefulness, so I don't see why we want to measure that and give it any name containing "useful" or "helpful".

However, if we want to keep the statistic based on votes, it should be named after what it is: Posts Popularity, or similar.

Unsure of one of the constraints you have as it contradicts a couple of badges, notably:

It's okay that it can't separate views that came before your answer, so long as it's very unusual that an answer posted way after the question on a popular question will count.

Which potentially goes against the Revival and Necromancer badges:

Necormancer
Answered a question more than 60 days later with score of 5 or more

This is perhaps quite a common case, as the Necromancer badge on SO appears to be the only Silver badge that has been awarded more times than its bronze equivalent:

More Silvers than Bronze

Actually, re-reading the text on those, they are only loosely related as the Bronze is only awarded if you're the First answer where as the Silver can be awarded for any answer.

  • Good point about the badges. As the queries for a proposed gold version show, it actually is hard to get upvotes for answer on ancient questions. – Nemo May 4 '15 at 8:51
  • This criterion doesn't go counter to the badges as much as reflect the contribution of the post to the thread's stats. If an old thread has 100K views and you post a good new answer, that answer didn't contribute to any of the previous views. Counting all views gives undo credit to the influence of your answer. – fixer1234 Jul 25 '17 at 17:28
  • @fixer1234 Indeed - I think my thoughts at the time were that if you did get a score of 5 or more for an old question, then it's possible that the question on it's own may not have helped that many people - if there were no answers on it, or not many good ones, then it's not really helping all those visitors other than to say "you're not alone" ;) – Zhaph - Ben Duguid Jul 26 '17 at 8:13

Ignoring zero-score questions might slightly penalize people who contribute a lot to niche categories. I think a lot of my answers have zero rep because a) they are the only answer, and b) they have only helped one or two people, and c) those people don't know how this site works so they don't vote or accept answers.

I don't have a concrete suggestion. Maybe include answers that are more than, say, two months old and have zero points and are the only answer, and the question is by someone who hasn't voted on anything. Or something like that. Maybe that number would be so small for almost everyone as to be useless.

  • Not sure. If I contribute to niche tags an high "people reached" score makes me go "oh no, massification!" Rewarding useful work on nice areas is rather the job of revival/necromancer/... badges (which should be expanded). – Nemo May 4 '15 at 8:53
  • Yeah, there's an argument for excluding only negatively scored answers. A zero score can just mean that nobody has yet found a good answer to be the solution. A negative score is more a statement that it is not a potentially useful contribution. – fixer1234 Jul 25 '17 at 17:35

The way search engines infer the "satisfaction" with their service is by measuring the time a user spends with a link they clicked. If the user comes back in less than 5 seconds (or another arbitrary cutoff, there is probably research on that) to click on the next search result or to start a new query, he counts as "unsatisfied".

Now this is not a perfect metric, but it's a part of how Google and Bing manage to show you stuff which not just has a high page rank, but is also relevant to your query. It's technically called dwell time and described in papers such as this one

I realize that Stack Exchange is probably not tracking this metric yet. But if it is, that number could certainly be included in the formula. And if not, it might have internal uses beside profile decoration, which could make it worthwhile to track.

Could you, going forward, keep a separate page view counter for each answer that only starts increasing once the answer is posted? For example, let's say I write a question that gets a million views in its first week of existance and a super-popular answer with a ton of upvotes. The question itself and the first few answers would all have their own view counts, stored separately, and all of these counters increase every time someone views the page.

Then, a year later, a new users arrives a posts an insightful answer. On the day it's posted, it also gets its own view counter that starts at 0. It also increases every time someone views the question, but starts out much lower than the views the older question and answers already have. Over the course of several months, the new answer gets five upvotes and the question is viewed a few hundred more times. Once the answer gets its fifth upvote, the hundreds of views it garnered are added to the user's "people helped" count.

class Question {
    int viewCount;
    List<Answer> answers;

    public logPageView() {
        viewCount++;
        for(answer in answers) {
            // Every time the question is viewed, all its answers also get views
            answer.logPageView();
        }
    }
}

class Answer {
    // Starts at zero when the answer is posted
    // Use this number in "people helped" statistic
    int viewCount;

    public logPageView() {
        // Called every time the associated question is viewed
        viewCount++;
    }
}
  • This could be coupled with knowledge of which answer is on top and when so as to avoid quick-fire answerers getting incorrect stats – David Wilkins Dec 5 '14 at 19:26
  • 1
    I was going to suggest something similar to this, but using a slightly different method. Simply store the question view count for when the answer is posted in the Posts table, and default all existing rows to 0. We over count existing posts, some to greater or lesser degree, but moving forward we get an accurate count. – MBraedley Dec 6 '14 at 3:35

Score > 0 AND

So, Unsung Heroes, aren't really heroes for that stat?

  • 1
    Otherwise they wouldn't be "unsung". – Nemo May 4 '15 at 8:54
  • @Nemo Good point! – falsarella May 4 '15 at 17:17
  • They'd still be heroes even if we sing a little. – fixer1234 Jul 25 '17 at 18:08

Sounds like a nice idea, but not really. The following reasons are in my mind now:

  • Easy to cheat as long as the views count is not recorded in a way to track how many visits are there in a specific day.
  • To my knowledge, web crawlers have a big effect on visits, adding more visits which will affect the result, making it more useless.
  • Reputation is accurate, and shows a glimpse of how helpful a person is.
  • In addition to reputation, we have badges, tags that every user is good at, vote count, activity, etc. All these items are accurate.

IMHO, an index composed of accurate data is much better, it doesn't have to be "people helped". My main reason behind this is the result I got from the query posted in data.se, it shows that I have helped 700,000 people in Travel.SE, I am sure I haven't helped even 10% of that number.

  • This is sort of true, but not as precise as implied. Reputation skyrockets for the wrong reasons, like a humorous answer on a HNQ that isn't at all useful, while an excellent, helpful answer on a niche question draws no attention. As a knowledge base, reputation is a clever metric, but doesn't always reflect the quality or usefulness of answers, and doesn't represent a consistent measure across questions. Views are mostly attracted by the question, and especially its title. Answer scores are driven more by the question than the answer. If a tree falls in the forest... – fixer1234 Jul 25 '17 at 17:48

I can only suggest an idea for the second big question

Big Question #2: What's a better name?

I've just spent the best of two hours reading all the connecting links, stories, anecdotes, tweets, etc. that populated and generated from Aaron Swartz's twenty-six years of life.

And although I couldn't possibly say I know what he must have been like in life, I am struck by the number of connections I have made in those two hours. Connections that are directly related to my own life, connections which shed a tiniest glimpse of light into what makes a man, woman, or young person passionate about something. I know nothing, nor am I particularly interested in the Python programming language, or anything to do with "cyber" technology, but nevertheless I have made discoveries following Aaron Swartz's links, tweets, and blog posts.

As a result I thought of the following:

- [10k] visitors connected

I think the verb connect can be interpreted on different levels

  1. Bring together or into contact so that a real or notional link is established.
  2. Provide or have a link or relationship with (someone or something).
  3. Link to a source of [information] supply.

I think this is an awesome idea. I'm also thinking it'll be a better indicator of a users value than rep count, if other stats like the ones alluded to do materialise.

However, I agree with Erwin that the name should be as unambiguous as possible.

Posts Viewed X Times is pretty unambiguous but I think it could be improved to Posts Visited X Times. Visited is even less ambiguous than Viewed in my opinion, as it is more suggestive that people revisit posts at a later date which I believe is closer to the truth.

I have personally visited many posts over and over again all across the network, some that I've answered, some I've posted and some that I learn from, copy code snippets or follow instructions from. I think from what I read of how page views are counted, that many of my own visits will have registered more than one page view, especially with different devices on different connections; as well as just revisiting a few days/weeks later.

So to summarise and reiterate:

       Posts Visited X Times

                would be closest to the truth.


On a different note, I think it would be cool if we had a choice between 3 or 4 options, adding a layer of personalisation / customisation to our profiles and making each slightly more unique in the process. (Someone suggested something like this in one of the billion comments.)

  • 1
    Number of times the post is visited or viewed has absolutely nothing to do with its quality ant this number is often even anticorrelated with the level of the post, at least on science sites: popular-level, basic posts are understandable to everybody and get therefore naturally much more views and visits, while research-level very high-level content can be understood and appreciated by experts and advanced students. Seems SE is now seriously optimizing all features of the software to favor popular content over research/expert-level contributions... – Dilaton Dec 5 '14 at 22:19

Just to add another drop of rain to the brainstorm, how about "wingspan"?

The more people you help, or the more that see your answers, or whatever, the bigger your wingspan.

Connotations with majestic birds -- the greater the wingspan the further they may fly. Connotations with airplanes -- roughly, the bigger the wingspan, the more they can carry (meaning either how many people you can carry, or how much support you can deliver to people)

Plus, it's ever so slightly whimsical.

  • Lemme guess: you are going to talk about bomb payload next, aren't you? :P 'Cause it rhymes better with reaching out and touching someone... – Deer Hunter Dec 5 '14 at 23:18

Use multiple formulae and take a mean (or mode, or median or whatever statistical method that seems appropriate)

I suggest use of more than one different formulae, each having different pros and cons. So one formula may be soft on those who give answers on popular questions very late, but other formula would have a opposite behaviour so they together should balance each other.

Stuff you may use

You could use various stats, some that come to my mind are:

  1. The more votes other answers with higher votes have, the lesser would be the reach (effect or helping power) of an answer. For example, an answer with five votes on a question having another answer with 50 votes will have lesser reach than an answer with five votes but, another answer having only seven votes.
  2. Taking a product of views and votes and some constant of Stack Exchange may be useful than just considering views. Also consider point #1.
  3. Ignore views in the first N days. You may ignore views in the first N days, because the initial views are generally by active users wandering around. They are not the ones who are helped. N will depend upon the specific site.

I'll add more things as I remember them.

  • I don't think any information about views as a function of time is stored in the database, only the total count. But they could be sampled after N days and stored. Does that fit into the current structure of the database? – Peter Mortensen Dec 7 '14 at 8:37

My problem with "people reached" is that there's a huge potential for double (or worse) counting here: if I keep posting, say, great Minecraft answers, then great, I'm helping a lot of people, but I'm probably not helping a distinct set of people each and every time. The prototype just sums views together.

It's more accurately a measure of the total number of times my contributions have been helpful.

That doesn't really help us get to name it properly, but I don't think it's appropriate to call the things that are being counted "people".

How about, perhaps: "metaphorical days saved." If a post of mine is helpful, it might've saved someone's day, metaphorically speaking. It's fine to double count days, and you can even scale it up seamlessly to count "metaphorical months", "metaphorical years", "metaphorical millenniums", etc.

And here's the killer feature: if you're feeling really cheeky, you can call it "metatime" ("metaday", "metayear", ...).

Some of what I am writing may seem or even clearly is horribly unfeasible – but I will address these issues in the end:


[…] nor can we count just the views on a question page that came in after a given answer was posted.

But you have the times of votes, which are correlated in time with the visits of a user that can vote, which in turn is correlated with overall views. Thus, as a first estimate you can estimate the temporal distribution of views by taking the temporal distribution of votes and just renormalise it by multiplying it with [number of views]/[number of votes]. Two tweaks to this:

  • Only regard the first votes of a user on a post instead of all votes, since this gives a better estimate of the visits of a user capable of voting.

  • The above estimate is certainly not perfect, e.g., I would expect the temporal mean of the distribution of actual views to be somewhat later than that of the distribution of votes, because the first votes come from SE power users and not from people having a similar problem. However, the error due to this is mostly systematic, i.e., it behaves similarly for all questions. Thus in order to correct for it, you only need to look at the actual temporal distribution of views (and votes) for a few questions, not all of them. Even if you do not have any data about this at all, it may suffice to start generating this data now, as the biggest discrepencies between those two distributions will likely occur shortly after the question is asked – or with other words: Those distributions have similar tails.


But, of course, you can deal without estimating the actual temporal distribution of views, because the stat needs (or even should) not estimate how many people viewed a question but to how many people it was useful to some extent. As already mentioned in some of the other answers, the only way we have to evaluate this are votes. Unfortunately, not every visitor is able to vote, but we can use the voters as proxies for the viewers. Given that you can only vote on what’s actually there, this also automatically addresses the problem of late answers to some extent. The resulting stat would be:

                               number of visitors 
score (per question) = ———————————————————————————————————  ×  upvotes
                        number of visitors that could vote

Some remarks on this:

  • An automatic feature is that this accounts for questions which are popular but lack good answers.

  • I chose only to include consider upvotes here, since I consider this the better measure for helpfulness to some extent (also considering downvotes would be more complicated since there may be a relevant number of visitors that can upvote but not downvote).

  • An obvious tweak would be to give more weight to upvotes by the asker as well as the accepted answer (with due consideration to the existence of askers who never upvoted or accepted anything).

  • Another take on this would be to ignore views, views by potential voters and votes happening shortly after the question was posted, with exception of votes and checkmarks by the asker. This way, you would get a better estimate for people who probably had a problem similar to the asker. This would again require a good estimate of the temporal distribution of visits and visits by users who could vote, as described in the first part of this answer. (If you want to go even further, exclude everything happening shortly after the question has been bumped.)

  • Of course, this requires that you have in some way recorded the number of visitors who could vote or at least have a good estimate for it, such as the number of logged-in visitors. If not, the next-best comparable score arguably is the one suggested by Mysticial.


While I am no database expert, I can guess that some of my (and other) suggestions may seem unfeasible due to requiring too many queries. However, it may very well be, that a fraction of an individual user’s post makes up for most of his score. In that case, it suffices to be precise for only those important posts.

I have an idea. Note, that I am not saying it is better, it just reflects the "number of people helped", most answerers seemed to desire. And I agree more and more! If you are not after that, you can stay with the existing system, which clearly has advantages, most notably including unregistered users. However, there are grave disadvantages, namely counting people who didn't even like your answer and counting people multiple times.

We want to measure people, right? Why bother with the inflated and unsatisfactory views? I would simply stick with registered users. My site, German Language Stack only has about 10K. The most straightforward way to determine, whether I ever helped someone, is to track, whether they ever upvoted any of my answers. You could make this strikter, if you wanted.

Anyway, the whole data for this is just a 10K x 10K matrix with binary entries. I don't see difficulty in storing that. Probably the matrix is sparse, then it's even easier to store.

You don't need to update this calculation life! You already store all activities. Once per month you check for the matrix. The non zero elements don't even need to be checked. If the matrix ever turns dense, because most people have upvoted each other, you just reverse the role of 1 and 0. I don't know, how the community works, so I can't offer good design ideas.

Personally, I am fine with omitting unregistered users, because I very much prefer a conservative estimate to an inflated one. But I understand that is purely a matter of taste. In this context, I would like to know what percentage of viewers the unregistered users represent. For the views to have some advantage, it would have to be high. I assume there are striking differences between answers with high and low votes. For votes below 50 I suspect most traffic is generated by the asker and a bunch of answerers that visit it 5-10 times each. A typical answer on my site has 2-3 answers. With my assumptions, that would generate 15-40 views from the askers and answerers.

Another thing is, if we are in favour of views, can't we at least separate IPs? That way, I don't give my answer ten views when I return to edit it, as I usually do. Still, some people would be counted twice and some rare groups of people would count as one, but It should be a great improvement. Even if I just stored the IPs for one day, aiming at something like "separate views" instead of "separate IPs".

Personally, I used to be really proud of this number. I thought "it probably won't be 14K" but let's say it's 1500". But now that I know how the statistic came about I can never look at it with pride again. Even a person who scrolled to Mary's answer thinking "hell, is Ludi's answer crap!", now counts as having been helped :(

Due to the intense disappointment this has caused me, I find myself contemplating even the following somewhat user unfriendly method for unregistered users: showing a long(!) preview of the answer and letting them click "continue". Though this can't measure helpfulness, it can measure interest at least.

  • A simple count distinct on the votes table should do. The data is already there. I am not sure though if I like your request or not. – Patrick Hofman Jul 18 '16 at 11:25
  • @PatrickHofman oh, it's not a request. Just a proposal. I think it better reflects the number of people helped, that many seemed to be after. But to me any variable is as good as any other, as long as I can read it. – Ludi Jul 18 '16 at 11:30
  • 3
    The major flaw here is the fact the great many people just view an answer, without having account, using it, and moving on. That's it. No trace left, and the only way to measure it is via the views count of the question. By limiting the stat to only existing users, you pretty much taking away the whole point in this "thread" here. – Shadow Wizard Jul 18 '16 at 11:33
  • @Sha on the other hand, viewed != useful, so total votes is a better measurement on the number of useful contributions. Still, that is not "people reached"... – Patrick Hofman Jul 18 '16 at 11:35
  • Reading this, everyone seemed to say " I want to measure people helped " but don't know why. So I threw in this. I'm fine with limiting to registered users, because I prefer a conservative estimate to an inflated one. But I understand all views. – Ludi Jul 18 '16 at 11:38
  • Because reputation already reflects the number of users helped. Rougly take your reputation and divide that by 10 (not accounting for accepts and downvotes). This statistic is much broader than that. – Patrick Hofman Jul 18 '16 at 11:40
  • @PatrickHofman well, that's the same problem backwards. I think it may be mostly the same people upvoting me. I don't think my votes are so distinctly distributed! – Ludi Jul 18 '16 at 11:44
  • Okay, good point @Ludi – Patrick Hofman Jul 18 '16 at 11:45

BogoMIPS

(or something similar--BogoHelps?) would be a better name, because it admits outright that it's bogus.

Jon Skeet has apparently "reached" nearly an order of magnitude more "people" than there are developers in the world.

Are extra terrestrials now reading SO?

  • It was never meant to be accurate. – Shadow Wizard Jan 6 '16 at 8:38
  • @ShadowWizard: Then give it a name that indicates as much. That's all I'm sayin'. – Flimzy Jan 6 '16 at 8:47
  • I am sure many tried to find better name and failed, I'm not expecting to be better than all of them. This works in 99% of the cases, I really don't think we should change the name to something complicated just for the rare cases where the number is way off. – Shadow Wizard Jan 6 '16 at 8:49
  • @ShadowWizard: This post asks for naming suggestions. That's what my answer provides. (I realize it's a year late, so new suggestions are unlikely to take hold without a massive upheaval of support, but I'm still offering a valid answer to the question). – Flimzy Jan 6 '16 at 9:07
  • I know, I never said otherwise or suggested to delete the answer, just explained why I disagree with such a name, and why I didn't come with answer of my own. – Shadow Wizard Jan 6 '16 at 9:11
  • @ShadowWizard: Fair enough. – Flimzy Jan 6 '16 at 9:18

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .