Chapter 16 Secondary analyses

If you have opted into our ChildProject framework, then there are a number of secondary analyses that become incredibly easy. We have thought about the most typical routines you need to engage in, and provide sample code or direct functions for all of them.

In particular, one thing you typically want to do is estimate the child-session-level metrics. For instance, one child was recorded for 8 hours one day and 16 hours another day. You don’t want to have the play-by-play, that is, when the child was vocalizing. But instead, you may be interested in finding out, in total, how much they vocalized in each of these recordings. Since the recordings lasted a different number of hours, you may want to get the recording length and estimate a number of vocalizations per hour instead. Getting all of these numbers is easy once your data are in the ChildProject format, and you can find example code for estimating these “metrics.”

Another frequently encountered issue is how accurate automated analyses are. We discussed in the Video on 15 that to this end, you may want to do human annotation. You will find an example of how to set up an annotation campaign in the Resources.

Moreover, once you’ve gotten some sections annotated, how do you estimate the reliability? ChildProject functions make it easy to calculate recall, precision, agreement, and other reliability metrics (such as error rates and correlations). Check the Resources section for an example on reliability.

Some of you have so much data that you may want the help of citizen scientists in the Zooniverse platform to get them annotated. We have an example project that shows you how to extract clips and cut them up into short sections, so they can be shared over Zooniverse, and then get back the annotations, analyze them, and integrate them with your dataset – this is the best practices for Zooniverse example.

Do you have other questions for analyses not covered above? Well, peruse the “show and tell” conversations in the ChildProject discussion board for some ideas, and feel free to create your own!