Is the Way Polls Calculate Percentages Correct?

[Lordmiller]

I posted this as a mod request over a year ago. . .a year later, and I have seen many polls with multiple voting allowed. I find there's never been a single instance where the person posting the poll actually cared what the "% of votes" were, it's always the "% of voters" that matters. . . .in instances where multiple votes per voter were allowed, the poll was still trying to measure "What percentage of members agree/disagree with this choice?"

That said, it makes me wonder if the choice to calculate the percentages as "Votes for Line Item / Total Votes" instead of "Votes for Line Item / Total Voters" is actually a mistake, rather than just contrary to my preference?

Does anyone out there ever run a poll where they care about the % of votes rather than the % of voters?

If not, would this then qualify as a (minor) bug / logic error?

Did a search on the boards and found nothing, so I'm posting this. I hope this is the right area. If someone already asked and got an answer on this one, a point in the right direction would be much appreciated.

Currently, if you set up a Poll where multiple options can be selected, you get strange percentile results.

For example:

If you set up a potential menu and ask people what items they would eat:

Steak
Fish
Chicken

And then allow up to 3 votes per voter.

You could get 50 responses

with an average of 2 votes per voter, for 100 votes.

, and the results would look like:

Steak     40      40%   
Fish       20      20%
Chicken  40      40%

Voters: 50

When in reality, steak is 80% yes, Fish is 40% yes and Chicken is 80% yes. The poll is displaying the % as the percentage of total votes in all categories, rather than the percentage of voters.

So my question is:

Is there any way for the polls to report the percentage results when multiple voting is enabled so that the percentages of result reflect the percentage of voters, instead of the percentage of votes?

[karlbenson]

I don't think this is a bug. Its just a different type of poll your after.

Essentially your treating each option as a Independant question.
So
Fish 80% of people checked it

Whereas in smf, its comparitively between the options (so the % can never exceed 100%)

You'll probably need a mod to do this.  If the mod request forum isn't happening for you, then consider posting a free/paid request in the help wanted forum to see if someone will do it for you as custom work.

[Lordmiller]

I get the logic behind it, and I did ask for a mod a year ago when I first noticed the effect of allowing more than one vote in polls.

But since then, I've noticed that there doesn't seem to ever be a context in which "% of votes" is a useful measurement, and since everyone looking at polls assumes it means "% of voters" you get questions from your members, or comments, like "There's something wrong when the winner only got 28% of the vote." when they actually got 56% of the voters. Which I've explained. . .a number of times.

I was more wondering if anyone out there is ever looking for % of votes, vs % of voters .

I so far have never come up with a context where % of votes would ever be a relevant metric, despite running a hundred or so polls. I wasn't sure, that's why I asked if anyone out there actually finds the current percentage calculations useful for polls that allow more than 1 vote per voter.

If I'm alone, and most people do find "% of votes" a useful metric, then I totally agree that I'm an odd duck, and I need a mod, since everything is working the way it should be, and my needs would be satisfied with an individual solution.

But if I'm not an oddball, and nobody finds "% of votes" a useful metric, then it's one of those situations where the code works perfectly as it's supposed to, but the result is flawed, or at least, non-useful.

I guess if it's not broken, it's not a bug, but if nobody actually finds it useful calculating that way, then it might be worth changing anyway?

[karlbenson]

Personally I would always use percentage of actual votes cast, rather than percentage of possible votes.
(Especially with polls with many options, I don't believe that when a user doesn't check a particular option, that he is casting a vote against it)

But I do think both could be said to be valid logics.
I'm not sure (but would be interested to know) the logic used by other forum software.
At a guess I would think they use the same logic as smf.

A mod would probably be the best way to go, at least initially.
As with most features, they start out as mods, and if they prove popular, then they might end up becoming core features.
(quick reply comes to mind)

[Lordmiller]

"% of votERs" is what people seem to intuitively expect the results to mean, based on questions and comments.

With 1 vote per voter, the "% of votes cast" and "% of voters" are equivalent, the fact that I only get confusion/complaints on multiple vote per voter polls suggests that the confusion is due to the fact that people seem to expect any poll result to give a result of "What percentage of the people who answered this poll chose this option?"

I'm kind of confused how "percentage of actual votes cast" is useful, but you prefer it, could you perhaps explain why/how that's helpful? Perhaps if I see how it has a useful application, it'll stop bothering me.

[SleePy]

I don't see this as a bug either. As Regex said, it is just a different option you are after.

[Lordmiller]

I'm not disputing that, I'm remaining open minded to the possibility that I'm quirky and most people find the current resuts satisfying.

I've run a bunch of multi-vote polls, and looked around at other SMF sites, and every instance of a poll where multiple votes are allowed, the percentages associated with the results seem misleading at worst, and non-usefuly counter intuitive at best. Perhaps I'm not looking at the right spots, and someone can point out a multi-vote-per-user poll where the percentages on the results were useful.

As a for instance, a "Game of the year" poll, allowing three votes per user, resulted in a winner with 18% of the votes, which lead to surly and aggrieved responses, until I pointed out that the winner had 54% of the voters and showed the math.

I can offer an example of these two polls:

http://www.rpgrm.com/rmsmf/index.php?topic=3691.0
http://www.rpgrm.com/rmsmf/index.php?topic=3732.0

Can anyone say the percentages associated with the results on those two polls are useful in any way? Those 2 polls offer a huge number of votes per user, but really, the percentages seem to go awry even at 2 votes per user.

I'm willing to accept that it's not a bug, and that I merely want a different option. But, I'd like to see an example of a multiple votes per voter poll in which the result percentages are in any way useful. Can anyone post a link to such a poll?

Now, I'm willing to accept it's not a bug if I can see such a poll. If you can't find a poll allowing multiple votes per voter where the result percentages are useful, are you willing to consider that it actually might be a bug?

[karlbenson]

The thing is LordMiller, I'd use ANY poll as an example of it.

I'd just recommend re-posting your mod request topic or bumping the list one.

[Lordmiller]

That would mean that looking at the two examples I posted links for, you can honestly say those percentages associated with the results are useful somehow? Could you explain in any more detail than "It's fine the way it is."

I'm not ranting, nor am I being rude or hostile, I'm just pointing out that the results seem off. If I'm simply mistaken, I'd expect it would be simple enough just to link to a poll and say "Look at this poll, examine the percentages associated with the results, they are useful because of X and Y and Z.". . .

If "ANY poll" would work as an example of how the current calculations are useful, it makes it even easier, you can just use the examples I posted and just simply explain to me how those percentage results are somehow useful.

I'd like to think I'm fairly intelligent, mostly reasonable, and I'm definitely willing to listen. . .explain it to me.

But, in the spirit of fairness, if you can't. . ..would you be willing to consider the possability I might have a point?

[karlbenson]

Hopefully my answer didn't read like a "Because thats the way it is..." answer.

Its just hard to quantity it what for me (and possibly most people) is a normal default poll behavior.

Like i said before, I ain't arguing that either are us are wrong, its just one is more widely used (as is used as default here).
And the other isn't (to the best of my knowledge). But both are still perfectly valid.

Just like elections, there are multiple ways of calculating these things.  For probably a majority of polls on a majority of smf forums, the functionality provided by smf is sufficient.  For the remainder of people/forums, we generally ask them to request mods (to provide that alternative/extended functionality, just like people who want a shop use SMFShop, those who want multiple polls per topic, use Multiple Poll mod)

There isn't much else I can say really.  I don't think there is anything else I can add in here as a bug report.
- We've taken your report.
- Determined its not a bug. But that what you want is alternative perfectly valid functionality.
- As we always do with 'feature/functionality requests'. Recommend to post in mod requests.

[Yahmez]

I agree with Lordmiller. Although the current way SMF calculates polls is not mathematically incorrect, it is not useful when you need one vote per choice rather than one vote divided equally among multiple choices. Which as far as I can figure is the only useful way to calculate a poll with multiple choices. If nothing else, how about a radio button to toggle for a different way of calculating polls with 'multiple choices' enabled? 

Example of what I'm looking for:

What did you plant this year?

4 total voters

Tomatoes  4 votes   100%
Peppers     1 vote    25%   
Corn         2 votes   50%
Onions      2 votes   50%

[karlbenson]

2.0 is feature frozen so it won't get it 2.0.

That it why you are best looking for someone to make it as a mod. if it proves successful, like other features, it would be considered.

[Lordmiller]

Mod request is out and pending. . .

This isn't an earth shattering issue, and I can definitely see why it's not a priority. . .but I just personally don't see the utility of the current percentage calculations for multiple voting. Shrug, perhaps I'm wrong. I think that a look around at statistical analysis and vote counting methodology in use in every other context shows that votes per choice/voters seems to be the method used in general and the SMF multiple vote counting method to be an outlayer, outside the norm.

That doesn't make it wrong, but it does make it different from generally accepted practice.

Technically, votes usually count the percentage of voters. Product surveys, opinion polls. . .all care about the "Percentage of responders" who selected a line item. . .i.e votes for a choice/voters. Product surveys, Parliamentary elections, opinion polls. . .every real world context in which multiple votes per voter are allowed count votes for a choice/total voters.

There is not a single real world context I can locate in which votes for a choice/total votes is ever actually done, when you're only allowed one vote per line item. It does mathmatically calculate perfectly I won't deny that, it just uses a vote tally method I've never seen in any other multiple voting context where only one vote is allowed per line item.

It's a "Unique" method. . .

There is one kind of poll that calculates votes for a choice / total votes. . . .

Whenever you take a customer satisfaction or review survey in which you're offered options ala:

Customer Service was: (-2) Terrible, (-1) poor, (0) average, (+1) good, (+2) Excellent

This kind of poll allows each voter to put in -2 to +2 votes. . .per line item.

Thus, by counting votes per item/total votes you actually get a usefull result that weights the results based on how strongly positive or negative the voters were.

There's no way in SMF, even if you allow multiple votes, to get in multiple votes per line item.

So the current vote count method does usefully support a voting method (weighted opinion polls allowing multiple votes per line item) that's not available as a voting option in SMF.

shrug. . .It's easy to present examples of how the current method presents less-than-clear and less-than-usefull results with multiple voting. . .and I've yet to see a single example given of a poll in which the current multiple vote counting method was considered superior or even useful. . . . .I'd like to see a link to an actual poll, allowing multiple votes, in which this counting method offered the best possible results.

I'll happily eat my crow, or my hat, when presented with a concrete example.

[Yahmez]

I'd like to see a link to an actual poll, allowing multiple votes, in which this counting method offered the best possible results.

I'll happily eat my crow, or my hat, when presented with a concrete example.

It will never happen, because it doesn't exist. I wish someone that knew coding could help me understand the math to code relation so I could make the edits myself.