Today was a great day. After a few weeks of data collection, I finally finished data collection for my project. For a while, I was worried I would run out of time. Around the end of May, I was having a lot of trouble finding novel subjects who had not seen the experiment already, and was repeatedly testing the same monkey (some as many as four times!).

This happens occasionally, especially with subadults without any significant markings. Presently, we have to exclude a monkey who has already seen the entirety of the test one time, so that our final dataset(s) only include each monkey a single time. This makes sense — we wouldn’t want one monkey’s behavior (especially if it was abnormal) to characterize the results of the entire study. Even though it is commonplace, it is incredibly frustrating because in addition to not being able to count the data, you also feel like an idiot (!) for not recognizing a monkey that you’ve already successfully tested.

In any case, a complete dataset means that the next phase of data analysis begins — coding! While ‘coding’ typically means writing hours of code into some program, coding for a comparative psychologist may mean that, or mean ‘behavioral coding’, during which we go back and watch all of the videos we collected during data collection. Let me explain.

During data collection, we filmed 80 videos, one for every successful subject. Each subject saw three trials to complete a session. The next step in my data collection is to go through these videos and clip each by trial into three separate 10-second clips. I then randomly rename the clips so that they are anonymized. This is an important step to ensure that when we go back and watch videos, we are not injecting any subconscious bias based on how we want the experiment to work out.

Then, two people go back and rewatch all of the anonymized clips and record when they think the subject is attending to the experiment. Formally, this is called the ‘looking time method’ [for a *great* review, see Reference 1], and can be used to measure either the subject’s surprise or the subject’s preference for the stimuli (depending on the experiment and your initial pre-experiment predictions). The two people have to reach an agreement on when the monkey is attending at more than 90% (this varies slightly by subject and by experiment) of the video. We call this the ‘reliability’ between the two videos, and calculate the number by calculating the correlation coefficient between the two individual sets of coding.

Finally, after all of this is completed, we are able to look at whether the trends we predicted are significantly visible in the data. In my experiment, I used the looking time method to measure the subjects’ surprise, so I will be looking at whether my subjects were more surprised in the unexpected condition as compared to the subjects in the expected condition. I’ll check back in when I know the results of the study, but as you can see here, it will probably take some time!

REFERENCES

[1] Winters, S., Dubuc, C., & Higham, J. P. (2015). Perspectives: The looking time experimental paradigm in studies of animal visual perception and cognition. Ethology, 121(7), 625-640.