Question about discrepancy in total minutes after collapsing Part 5 averages in Stata

[viviane2403]

Hello everyone,

I am using the GGIR package to process accelerometer data, and I have a question that came up during post-processing in Stata.

We processed the data in R using GGIR and obtained the outputs from Parts 2 and 5. Then, we used the Day Summary file from Part 5 to apply quality criteria in Stata and collapsed the averages per participant.

However, we noticed that even after applying the quality criteria, the collapsed daily averages in Stata do not sum up to 1,440 minutes per day, as expected to represent 24 hours. There is a loss of minutes, especially in the light physical activity and moderate-to-vigorous physical activity variables. Sleep time and inactivity remain unchanged.

In general, the total daily time ends up around 1,300 to 1,400 minutes, but never reaches 1,440. This raised a concern because the outputs generated by GGIR, such as the Personal Summary and Day Summary, seem to correctly total 1,440 minutes.

Additionally, when collapsing the averages in Stata, we did not apply any weighting by day type (e.g., weekdays vs. weekends). Could this lack of weighting explain the discrepancy? Is weighting necessary in this context, and would failing to do so introduce any bias?

Our main question is: are the results we obtained in Stata incorrect? Is there an explanation for the difference between the collapsed averages we calculated using Part 5 (Day Summary) and the values shown in the Personal Summary?

We would really appreciate any guidance on the best approach in this case.

Kind regards,
Viviane Nogueira

[tarensanders]

I can think of two possible reasons (and if you post your code I could probably tell you which it is):

1. You're using a timewindow which is not midnight-midnight. The other timewindows are based on when participants either wake or fall asleep, and these won't sum to exactly 24 hours.
2. You're including incomplete days in your average.