Ten fold

whuber

03:37

@mbq That Stata question raises an issue that merits discussion. It would certainly receive a swift, authoritative, correct answer on the Stata listserver. Most other stats software has comparable obvious places to go for software-specific questions. Do we really want to get involved in handling these kinds of questions? One point of view is that we should stick to what we're uniquely good at and refer questions that are better fielded elsewhere to appropriate sites.

@mbq Nice graphics; great work. I disagree with one interpretation, though: there is a positive association between high reputation and points per response once you recognize the heteroscedastic structure: low rep = few answers = high variance. I think Srikant once alluded to the possibility of such an association as a "reputation effect."

ars

05:39

@whuber: I think one general argument is as follows. There ought to be a place on SE to ask useful question. For statistical computing, we have SO and CV. You're highly unlikely to find Stata programmers on SO, so CV is a natural place to ask these questions and we should be welcoming. A possible benefit is that it leads to interesting statistical "discussions". Of course, you need to accept the premise that there ought to be an SE for all such questions.

Another argument made on meta is that it's hard to figure out what to reject in advance. The dividing line is blurred and rather than have questions turn into a comment-fest of users deciding if a question makes the cut, we should just adopt an inclusive approach.

@whuber: nice observation on the variances of rep/post.

mbq

08:48

@whuber I agree; I had in mind that there are no people that earned a lot of reputation from few very high-voted answers. Heteroscedaicity on that plot is also boosted with the log scale.

whuber

14:30

@ars The gentle approach is to accept whatever does come our way but to provide clear guidance to help people post their questions in the right forum in the first place. For instance, we could have a "software" section in the FAQ with links to the R, Stata, SAS, etc. forums.

@mbq I don't follow you. The heteroscedasticity primarily affects the vertical direction, which has no log scale. (Indeed, it would be good to use a square root scale on the vertical axis because that should approximately symmetrize the variation, which has a positive skew.)

csgillespie

16:32

Following on from @mbq's analysis, I thought it might be interesting to look at differences between the different beta SO sites. Use a very hacky R script I download the number of votes, users, questions and view for each site. Then constructed a PCA plot.

mbq

@whuber But log expands the low reputation region when this problem is mostly present, so artificially "boosts" its significance.

csgillespie

Other than math, webapps, programming most sites are fairly similar. Programmers is different from other sites because they have a very high vote to user ratio

Math and webapps are different because they have a large number of questions.

mbq

Interesting result; this is non-intuitive but this shows that there is something like "median SE site" in sense of popularity.

chl

16:55

@csgillespie Interesting. What part of variance is accounted for by the first axis?

whuber

17:10

@mbq There is a significant problem with the fact that some users start with rep of 1 and others with rep of 101. For this reason the data will just be a mess in the low regions. That works against any graphic that would expand detail in this range. IMO we don't have any reliable information until reputation substantially exceeds 200.

csgillespie

@whuber I've got data sets where I've downloaded all users with the aim of discarding users < 150 (say). I'm still messing about with that though.

Shane

17:25

Argh. Jeff Atwood just denied our request to move from stats.SE to data.SE...

And Robert Cartaino says that we should go live on Thursday with the new design.

We will have the name CrossValidated, but will still be on stats.SE...

csgillespie

17:42

@chl I've added all the details at this webpage

chl

@csgillespie Thanks for this. I'll look at it.

Shane

@csgillespie Neat. Thanks for posting.

ars

19:36

@whuber: I find the second argument I listed persuasive because I don't like the comment-fest and rush to close questions. But your "gentle approach" is a nice compromise that I would favor.

@shane: D'oh! At least we get CrossValidated.

@csgillespie: Very nice. I wish we could get some of that math mojo.

Jin

20:10

hi guys

Shane

20:55

Hi @Jin.

This is the new room?

Jin

Hey @Shane

I'm designing the theme for the CrossValidation site this week.

I'm glad you guys came up with some nice logo ideas already

Shane

Great. We're excited for it.

Any questions at this point, beyond what you're seeing on Meta?

BTW. I think the name is "CrossValidated", not "CrossValidation".

Jin

right sorry. meant CrossValidated

Shane

We've been trying to use the theory behind cross validation as inspiration for the logo.

Jin

for now is it OK if I go with

I may tweak the coloring some, but the idea of the logo will stay the same

Shane

20:58

That's my preference. The scatter plot seems to be the crowd favorite, but it would be hard to design a logo like that.

Yes, that communicates the site idea quite well.

Jin

or, do the colors mean anything significant in the statistics world?

Shane

Not that I'm aware of, no.

@whuber Those colors are meaningless, right?

@mbq You can confirm this too...

Jin

I'm going for a clean minimalist theme, so it won't be graphics heavy like some of the sites I've done.

Shane

@Jin are you planning on being in here much during the design process? If so, I'll put up a notice to that effect on the site.

Perfect.

Jin

@Shane you mean in this room?

Shane

21:00

I really liked @mbq's idea about having the normal distribution for voting icons.

Yes, this room.

Jin

yes I'll be here on and off.

Shane

You could have the scatterplot with a best-fit line as the accepted icon.

whuber

@shane Who am I to tell what meaning colors have? :-) Importantly, they will work for most forms of color blindness (which afflict about 6% of Western populations), so that's ok.

Shane

Ok. That's great.

Jin

some little details I think that will spice the site up some:

graphics for accepted answers background on the homepage/question page

i remember someone suggested a curve, or was that for the voting arrows?

Shane

21:02

8

A: What should our logo and site design look like?

Gaussian-like idea for "voting component", so that it has a statistical interpretation:

Someone else had suggested the same for accepted, but maybe a scatterplot with a best-fit line as the accepted would be a good compromise.

A play on this for the accepted:

5

A: What should our logo and site design look like?

For the logo, another option would be to show a scatter plot with a best-fit line through it, possibly using a non-linear relationship (like loess) as in this example. This has nothing to do with cross-validation, but it would certainly communicate the purpose of the site. I prefer the other ...

Jin

@Shane I'm afraid a scatterplot would be too noisy for the accepted answer bg

as I've found out with some other sites.

Shane

Got it. Well, there were other ideas too, like this one.

whuber

@csgillespie Consider discarding users with fewer than k answers + j questions where k + j/4 is around 5-10 or so. That side-steps the 1 vs 101 starting rep issue. Also, for analysis, one could randomly combine data from small groups of low-scoring users in order to achieve approximate equality of variance with the high-scoring users.

@ars I am persuaded by your second argument too.

Jin

@shane what about this?

i think the background for accepted answers should be 1 shade, so it's very easy to read the numbers. I tried to do something fancy with the Math site, and got pretty negative feedback about it

Shane

Looks great to me.

whuber

21:11

I like this spare simple approach. Less is more. Statisticians usually want to communicate numbers, not decorate them.

Jin

@whuber I think so too. so I want to find a fine balance of decoration vs boring. so my approach is go for a clean look

i like this

Shane

Yes, I thought that was neat.

Jin

but it's not scalable since the number digits are dynamic

but since we'll have no more than 4 digits there, i may pull it off

i'd just have to right align the numbers so it flushes against the right edge

btw do you guys have any statistics forum/blog/sites URLs that I can take a look at?

whuber

@jin I guess if we ever need five digits you'll just have to redesign the site :-).

Shane

r-bloggers.com is a blog aggregator for the R language for statistical computing.

whuber

21:15

Andrew Gelman's blog is at stat.columbia.edu/~gelman/blog

Shane

10

Q: What statistical blogs would you recommend?

I currently subscribe to Andrew Gelman's blog Christian Roberts' blog Darren Wilkinson's blog What other research-level statistical blogs are there?

blog

Gelman's is probably the most popular.

Jin

cool thanks

whuber

The ASA is trying to get blogs started: magazine.amstat.org/blog/2010/09/01/blogstatistics .

Jin

for tag/badge treatment, what do you think would be appropraite for the stats site?

especially the gold/silver/bronze icons

Shane

8

Q: What is your favorite data visualization blog?

What is the best blog on data visualization? I'm making this question a community wiki since it is highly subjective. Please limit each answer to one link.

data-visualization blog best-of

This can also be useful/relevant.

The site has a broader scope that just statistics...

Jin

21:19

right. i'm a fan of flowingdata and info is beautiful sites.

@shane do you think some visual elements that convey "infograph" would be appropriate for this site?

Shane

Absolutely.

Data visualization is within the scope, so anything that communicates that will also be applicable.

stats.stackexchange.com/faq

Jin

@shane great! now this opens a lot of possiblities. (i won't go crazy though)

Shane

Notice that it's the third item listed.

Sounds good.

That's actually a little more neutral for the community than just doing statistics stuff (there are camps around other areas)...

Jin

great. also, what do you guys have in mind for the 404/captcha/error images?

Shane

Hmmm...we hadn't raised that question yet.

Let's start a meta topic on it.

Is it easy to change down the road?

We could do an infographic that has data on-top of a map, and say something like "are you lost?"

Jin

21:25

of course. i can always use some place holder for those images first, then change to something better later.

Shane

Great. I'll start up a thread and we'll see what comes out.

I have to take off. Will check in on this again later.

Thanks for all the effort!

Jin

thanks for the tips!

mbq

Seems I've missed an interesting discussion

Still I agree with what you've justified. And I'm glad we are going into minimalism.

Jin

hi @mbq

I'm glad I learned about the infograph aspect today. will give me more design ideas.

mbq

@Jin Oh, yes, hi.

Hope you'll find something interesting; do you have any ideas for the colour scheme?

Jin

21:34

@mbq I was going for a white or off white

with possible faint grid lines. and use some primary colors for accents. do you have anything in mind?

mbq

Looks ok; indeed I asked because we couldn't find any good idea.

Jin

i'm also toying with the idea of assigning icons to each of the main navs (questions/tags/badges etc). normally if the site is minimal, i prefer to balance it with some graphics elements in the header

www.webapps.com for example.

but if i do use icons for stats site, it'd be the flat 2d minimalist styled icons, not the web2.0 type

mbq

Any idea for badges?

I thought they could follow logo;

Jin

well there are 2 parts to the badges. one is the small icon, as you see on the top bar

and another is the icon on a badge background, as seem on the badge page, or homepage's sidebar

i'm not sure what would be most appropriate for the stats site yet...

mbq

The shape is the same isn't it?

Jin

21:39

the small icon portion has to be fairly small, otherwise the whole badge would be huge.

@mbq yes.

well there are some exceptions I've done. for example the badges on the cooking site:

cooking.stackexchange.com/badges

mbq

This two corner rounded rectangle is something that scales well enough.

Jin

oops i meant the webapps site: webapps.stackexchange.com/badges

@mbq ok i'll go with that.

mbq

@Jin So I thought.

So this completes my wishlist (-;

Jin

btw, i had an idea for the badges

see what you think

do they look like distribution chart to you?

this is different from the badges for the other sites, where all the gold/silver/bronze icons are in the same position.

or do you think this looks too disorientating, especially when viewed all on the /badges page

the 2 roudned corner square would do great as the badge icons too.

mbq

Frankly I think it would look better with tiny boxplots on the same position.

Jin

21:50

ok

mbq

@Jin Yet the logo-like is my favorite.

Jin

is that what you had in mind?

i shaped the badge background to be similar to the icon.

mbq

Great.

Transcript for