Summing locomotor activity data into “bins”: How to avoid artifact in spectral analysis

Abstract

Sequences of single events form a time series with characteristics quite different from discretely sampled continuous processes. The events of the discontinuous sequence are summed over fixed sampling intervals called bins, whose duration thus determines the sampling frequency. The rate of the process or amplitude of the signal is estimated by the number of events per unit time. Techniques normally used to analyze continuous processes are commonly employed on these transformed series. There are several potential dangers in this approach. Using a computer model, we explored the effects of binning on subsequent data analsysis. The binning process acts as a digital filter which passes low frequencies. The numerically approximated transfer function is irregular, and interacts with the data to pass some frequencies preferentially. Spectral analysis is not enhanced by using arbitrarily small sampling intervals, as it would be with a continuous process. The result of employing excessively small bins is loss of signal strength, while the result of employing excessively large bins is the inability to discern short periods in the record. We have developed “rules of thumb”; to choose appropriate bin size for locomotor activity records, and we explore the effects of binning upon signals with known properties. We have also identified sources of artifact in the binning process.

Abstracting keywords:

• Spectral analysis
• bin size
• digital filter
• biological rhythm
• data analysis

Notes

Author to whom correspondence should be addressed: Dept. of Zoology, University of Maine, Orono, Maine 04469, BITNET DOWSE@MAINE