Ellen Collins, the primary statistician for the study, posted the most statistically significant results to their blog (see Final Blog). Unlike many statisticians, she has done an excellent job in presentation. I was pleased that she posted effect sizes rather than raw values. This is important because effect sizes are so much more informative than the values, even when the latter is presented with statistical significance. Significance by itself only tells you that the differences between the groups you're comparing are not likely due to chance alone. Sometimes this is because the differences are quite large - sometimes, particularly with very large sample sizes, the differences may be rather small. By reporting the differences as "effect sizes", which takes into consideration variation within the groups, we, the readers, get a better understanding of how big that difference actually is.
It was also quite nice that she used colors to highlight the size and direction of the effect sizes. This makes the charts so much quicker to read and easier to understand. Here is one example:
Without this use of colors, she would have had to add more text to indicate the direction (the sign is irrelevant to direction) and it would have taken the reader more time to make sense of the values. Finally, she does not include data for relationships that showed no statistically significant correlation. This saves space and the time of the reader to skim through. For a peer-reviewed journal article, though, she would likely need to show these values so that the reader would be able to verify this information.
Now to the results - Like all good statisticians, she starts with the basics - associations of demographics with the library usage. In her discussions, she is careful to detail some limitations to the results and analysis. One of these is that some groupings were too small to analyze because they couldn't protect student identities and such analyses wouldn't produce valid conclusions. Another is that the effect sizes of the associations of demographic variables with library usage, while statistically significant, were not very large (she thoughtfully included descriptions of ranges of effect sizes). It's important to look at demographics first because these are factors that are beyond the control of the library, and if the effects were large, they could confound the relationship of associations between usage and outcomes. I would have liked to have seen, however, similar analyses of demographics with educational outcomes.
After determining that demographics have very little effect on library usage, she then looked at disciplines in both broad and more narrow groupings (see Findings Post 2). This has the most substantive, although admittedly less surprising, results. Comparing all the other broad "subject groups" to social sciences (the "majority" set of users), those in the Arts, as well as Computer Science & Engineering fields used the library significantly less in nearly all measures of usage. Interestingly, those in the Arts showed stronger effects of lower usage in the electronic resources measures. Now, the Arts includes video and music, and there are a growing number of streaming services that would support these fields. I glanced at the Huddersfield Library's resources for music and noticed only a few of these streaming services. These results could provide the support needed for additional funding. While it is not surprising that CS&E students did not borrow as many books from the library as those in the social sciences, our data of ebook usage shows that books these fields had greater usage than all the other STEM titles. While we cannot determine who used these books, many of the titles are pretty specialized.
I'll have more about these results in another post...
No comments:
Post a Comment