Lies, damned lies, and statistics

[danbliss]

Ok, I am relatively new to the site, but I am having a lot of fun here (although it was depressing that I got

handed a few of 3/3's on a couple of my best photos, killing the averages of those photos. Alas...).

 

I know the site started out as Philip's web site, and he is a MIT computer scientist, so I am guessing that

while this primarily an artistic site, that there is, underneath, something of an appreciation for scientific

thinking. That is a long winded way of introducing the following question: Is there some sort of analysis

of the contributions that is available to the average user. Clearly the numbers are huge. I thought there

might be some interesting social significance to the flow of artistic images. If you hadn't guessed already,

I'm a researcher at MIT. Anyway, I thought it might be fun to analyze the data.

 

Just thinking out loud,

Dan

[afs760bf]

I think first you would have to have data. And secondly, I think you would have to have reliable data. Determining the social significance of a depressing 3/3 vs a non-depressing 7/7 would certainly be a challenge.

 

Well, it's a great idea, Dan. I think you should pursue it.

[danbliss]

Well, I am guessing that understanding the 3/3's is beyond my skill (assuming that it is not

the fault of my photographs :-) ), but I thought the rest would be fun.

[wood]

Dan, what do you mean by "contributions"? Obviously, Philip or Jin or someone familiar with the database would have to provide the data, but would they be willing?

[philg]

Just say what questions you think are interesting to answer and what data you want to answer the question and we can probably provide it. The whole rating thing started out as a one-morning "fun thing to add" and turned into a nightmare with people registering fake accounts, building rating robots, etc.

[joshroot]

I just had a great vision in my head of this overworked slave robot chained to a desk and forced to rate photo.net images all day long. Heck, if I were that robot, I'd probably rate everything really low as well.

[bobatkins]

I believe the rating robot is, in fact, related to the well known R2D2. It's name is A303.

 

While the ratings may be a mess, I believe that they are probably the feature of the site that currently attracts most users.

[davidmccracken]

It is clear from most of the people who contribute to the site feedback forum that ratings are NOT what attract people to the site. What attracts people to the site are critiques, after all that is why we submit photographs for 'critique.' Perhaps it would be interesting if there was a choice to 'Submit for Rating Only' to see how many would actually do this!

[mike dixon]

<i>It is clear from most of the people who contribute to the site feedback forum that ratings are NOT what attract people to the site. </i><P>

I'm not sure of the exact statistic, but people who contribute to the site feedback forum probably make up less than a tenth of a percent of the people who participate in the rating system. Any conclusions you draw about what the vast silent majority thinks based on the views of the relatively few people who participate here have an extremely weak foundation.

[root]

It isn't about rates or critiques - or at least it shouldn't be. It should be about views. If the images you create have any value consistent with the goals of the site, then they should get more views. We seem to place value on knowledgeable people who can answer forum questions but allow the lowest common denominator tastes to determine what images are visible. This practice ultimately determines who will consider it worth their while to upload images to this site. Given the overall traffic on PN, it should be able to figure about a way to attract a much greater variety of photographers to upload their images here.

 

Some of those potential uploaders are already here, yet don't participate in the gallery games for obvious reasons. I'm surprised that no one seems to be interested in coming with ways to make it better.

[danbliss]

Ratings:

It is interesting that we got into the discussion of the ratings. While I was not clear,

because I am still formulating what questions I'd like to ask, I was not thinking about

ratings. I mentioned the 3/3 as a joke, but it is interesting how it does resonate with all of

us. It is funny how we are drawn to being rated. I am not immune to this silliness.

 

Other questions:

When I originally posted, the questions I had in mind were related to the rate of postings,

and the origins of the postings. I assume we could use the IP addresses to determine

approximate source location. We would also have to define what is meant by a postings.

Are we talking about uploading a photograph, or requesting a critique. I thought it would

be interesting to track the diurnal variation of contributions. I had the vision of a globe

with contribution density as a function of time plotted on it. I am sure that there are

better methods for displaying the data.

 

By analogy on a very small scale, I was working with several people on program a

communication simulator. There were a lot of lines of code. We used CVS (the software,

not the drug store) as our code management software. One of the guys on the team found

this software that analyzed submissions to the CVS repository. It made an infinite number

of plots. It was hilarious. You could track contributions by time of day and day of week.

One guy never contributed after 5pm. Another never contributed before noon. You could

see the frequency of contributions increase as we got closes to our deadline. I thought the

plots were so amusing that I used a few of them in my briefing to the sponsor.

 

Fun facts:

Given the size of the data set, I am willing to bet there are a large number of fun facts that

can be teased from it. If one looks at all the temporal, spatial, and stylistic correlations,

there may be some fun relationships that we didn't expect. Forgive the technobabble, but

I thought we might start by taking a representative data matrices, and look at the singular

value structure to see if there were any interesting patters. The dominant singular vectors

might indicate interesting correlations between whatever variables we were looking at.

Anyway, the literature is full of "data mining" techniques. From the perspective of Philip

and the rest, it might not be worth the effort. I would not blame them. But, I was just

thinking about it. Who knows, maybe we could find a graduate student who might be

interested in it. There must be some student from the EE/CS/Media lab with a signal

processing/data analysis bent. I'd work with them.

 

Privacy:

To be clear, we would have to take measures to assure that there wasn't even the

appearance of violating anyone's privacy. I say that knowing that it means that I'll never

find out who gave me the 3/3's. :-)