I’ve mostly been analyzing the data here by looking at how many articles (or how many expected articles) are in a topic. But there’s a case to be made for using pages rather than articles as the basic measure. To give a sense of how much this matters, here are the topics with the longest and shortest average lengths.19
|Origins and purposes||11.2|
|Propositions and implications||11.3|
The difference between the longest and the shortest is almost a 2:1 ratio. So if we measured things by pages, it would make some changes. But in most cases, this difference won’t show up on the graphs I’ve displayed so far. That’s because most of the differences are screened by the changes in average article lengths over time. That is, most of the changes are due to the fact that the topics appear at different times in the data, and that the norm for article lengths change over time.
There are some exceptions to this. The truth topic has a lot of shortish articles appearing in recent times, when the article lengths have been getting longer. Though even there something weird happens. Many of these articles are in Analysis, and they are very long by Analysis standards, but short overall. Still, it’s worth looking at just how much article lengths have changed to provide a sense of how much these changes in length trends are driving the tables in this section.
Note I’m using weighted, or expected, articles here, so for each topic I’m summing over all articles the probability of the article being in that topic times the length of the article and dividing by the expected number of articles in the topic.↩︎