What's the best time to post a question to get the most answers/views?

[Earthling]

@Shalmanese

Are you saying you gathered all this data by hand? A simple spider would trivially be able to collect the data I need

Yes, I gathered the data by hand. It may be a trivial task for a spider, but since I’ve never used or written one before (never needed to), the thought didn’t cross my mind. I will say, though, that collecting basic thread information can be made somewhat bearable by manipulating the display parameters to show more than 50 threads per page, and saving the HTML from each page for (semi)automated parsing later.

How many threads are active but not started last year? I can’t imagine it would be that many.

Out of the 69,117 threads active last year, 1,154 had earlier start dates. Out of these, about 110 were started in the year 2000. That’s a pretty small percentage, but it’s evidently enough to pull the correlations significantly downward.

How are you measuring lifespan? The common method is to go from something like the 10th percentile to the 90th percentile range to get rid off all the outliers, I suspect this is what’s fouling up your results.

Lifespan, as I’ve used it here, is simply the number of days between a thread’s OP and its last post. Discarding outliers doesn’t really change the correlations that much, though. Roughly 90% of all threads “live” for one week or less, and the correlation between lifespan and either views or replies in this range is only around 0.55 to 0.6 or so.

Hrmm… it’s interesting that nearly all threads are restricted to within a very distinct band but seems fairly evenly distributed within that band. Could you post a zoomed graph at around 200x5000? Maybe it’s just the dot’s obscuring the true data. Alternatively, could you post the xls file? I would be very tentative about the bifurcation hypothesis, it doesn’t look like there is enough data to confirm it. Of course, you get the “Is my vagina normal” threads that generate an inordinate amount of views.

As you wish. All this work is based on the same Excel file I’ve linked to above, ThreadLife.xls (3.5 MB .ZIP file).

Incidentally, one reason I had to throw away outliers for my original scatter plot was because the most-viewed thread (the LotR one, of course) has about 10x the hits as the second-place thread in the sample, and plotting it would just push every other data point into a tiny corner. But then I realized that this is exactly what a log-log chart is good for, so here it is, with all outliers intact (it’s still sampled down from the 69k+ original threads, though, because Excel can only handle up to 32,000 data pairs on a x-y chart). The general shape tapers to a point, and I haven’t thought about whether this has any meaning or not.

Hm, have we moved away from Chairman Pow’s original question?

[Chronos]

This suggests that, in general, the number of views and replies to a thread are strongly linked – and I believe that’s just intuitively sensible – but the few ancient threads that are somehow still active throw things all out of whack.

And this, of course, would be because the older versions of the software didn’t track number of views, so any threads from before the time of that conversion would have started off with zero views, even with many posts. Can you confirm that the older threads tend to have a lower view to post ratio?

Also, on your log-log views vs. posts graph, are you counting replies or posts (=replies plus 1 OP)? Your label says “replies”, but that would shuffle completely ignored threads out of the range of a log-log plot. It might be better to use posts, so the zero-reply (and therefore 1 post) threads would still show up.

[Earthling]

@Chronos

Can you confirm that the older threads tend to have a lower view to post ratio?

From last year’s data:
Threads started in 2000 and before: View/Reply ratio = 40.1
2001: 39.3
2002: 50.9
2004: 23.2

I don’t think these numbers are meaningful, though, since, as pointed out above, most threads only last a week or less, and this data set (threads active within the last year) just doesn’t capture the majority of the early threads.

Also, on your log-log views vs. posts graph, are you counting replies or posts (=replies plus 1 OP)? Your label says “replies”, but that would shuffle completely ignored threads out of the range of a log-log plot. It might be better to use posts, so the zero-reply (and therefore 1 post) threads would still show up.

On that graph, I’m counting replies. I had actually generated another plot with posts instead of replies, but somehow didn’t save it. (D’oh!) Anyway, here it is now.

From looking at the no-reply threads again, and somewhat pertinent to the OP, I noticed two threads with inordinately high hits that were missed by the sample (one is around 7500, and the other, 3500). These threads remain open, but no one has yet bothered to post to them. Both were started by banned users, though, so I don’t think that’s a tactic the OP would want to employ.