Evaluation of data loggers, sampling intervals, and editing techniques for measuring the lying behavior of dairy cattle

Lying behavior in dairy cattle can provide insight into how cows interact with their environment. Although lying behavior is a useful indicator of cow comfort, it can be time consuming to measure. In response to these time constraints, using data loggers to automate behavioral recording has become increasingly common. We tested the accuracy of the Onset Pendant G data logger (Onset Computer Corporation, Bourne, MA) for measuring lying behavior in dairy cattle (n = 24 cows; 12 in each of 2 experiments). Cows wore the logger on the lateral (experiment 1) or medial (experiment 2) side of the hind leg above the metatarsophalangeal joint. Loggers recorded behavior at 4 sampling intervals (6, 30, 60, and 300 s) for at least 1.5 d. Data were smoothed using 3 editing methods to examine the effects of short, potentially erroneous readings. For this purpose, Microsoft Excel macros (Microsoft Corp., Redmond, WA) converted readings (i.e., lying events bordered by standing or vice versa) occurring singly or in consecutive runs of <= 2 or <= 6. Behavior was simultaneously recorded with digital video equipment. The logger accurately measured lying and standing. For example, predictability, sensitivity, and specificity were >99% using 30-s sampling and the single-event filter compared with continuously scored video recordings. The 6- and 30-s sampling intervals were comparable for all aspects of lying behavior when short events were filtered from the data set. Estimates of lying time generated from the 300-s interval unfiltered regimen were positively related (R-2 >= 0.99) to estimates of lying time from video, but this sampling regimen overestimated the number of lying bouts. This is likely because short standing and lying bouts were missed (12 and 34% of lying and standing bouts were <300 s in experiment 1 and 2, respectively). In summary, the data logger accurately measured all aspects of lying behavior when the sampling interval was <= 30 s and when short readings of lying and standing were filtered from the data set.