I have done 40-odd radio interviews about The Shangri-La Diet. A few of them noticeably increased the number of visitors to the SLD forums. The forums software measures number of visitors with a statistic called most online: the maximum number of simultaneous visitors to the forums, computing the maximum over one day. Until recently, the interview with the most impact was with Dennis Prager (May 2006). For the five days before the interview, most online = 54 40 43 53 48. On the day of the interview, most online = 231. For the five following days, most online = 72 82 44 52 74. This was the general pattern: An increase caused by an interview lasted one day or a little more.
On June 28, I was a guest on The David Lawrence Show. This one was different. For the 10 days before the show, most online =
52 38 35 44 47 44 48 40 42 44
On the day of the show (live), most online = 148. On the following days,
137 119 164 281 88 95 207 89 130 128 164 96
A much longer increase than usual. And the maximum value so far (281) happened days after the show. To make it clear how unusual this is, here is a graph of most online for the whole history of the forums.
Why was the David Lawrence interview so different? I can think of four possibilities: 1. The SLD forums have become more persuasive. Because they are more persuasive, new visitors come more than once. 2. David Lawrence listeners were more likely to be persuaded by the forums. 3. The David Lawrence Show gets lots of listeners via the website, where the interview can be downloaded. These downloads happened over many days, so the effect of the interview was spread out. 4. Something was different about the interview itself.
Another measure of forums activity is provided by the hosting service: unique visitors per day. This is more interesting, of course, but harder for me to record. Only recently did I start recording it. Here is a graph of visitors per day:
These data support the idea that the David Lawrence interview had a long-lasting effect, yes — but why the decrease in variability?
Podcast: Play in new window | Download
It looks like you have this plotted on some sort of log scale. If you take it off that scale, my guess is there isn’t much decrease in variability.
Yes, it is plotted on a log scale. When counts are plotted on square root or log scales, the variability is usually constant with changes in level — one of the big reasons for the transformation. This data violates that expectation.
If it were plotted on a “raw” (untransformed) scale, the reduction in variability would still be obvious because the pre-interview values cover such a wide range.