New approaches to selecting a scan-sampling method for chicken behavioral observations and their practical implications

Abstract

The use of the scan-sampling method, especially when a large amount of data is collected, has become widespread in behavioral studies. However, there are no specific guidelines regarding the choice of the sampling interval in different conditions. Thus, establishing a standard approach for video analysis represents an important step forward within the scientific community. In the present work, we hypothesized that the length of the sampling interval could influence the results of chicken behavioral study, for which we evaluated the reliability, accuracy, and validity of three different sampling intervals (10, 15 and 30 min). The Bland-Altman test was proposed as an innovative approach to compare sampling intervals and support researcher choices. Moreover, these sampling intervals were applied to compare the behavior of 4 chicken genotypes kept under free-range conditions. The Bland-Altman plots suggested that sampling intervals greater than 10 min lead to biases in the estimation of rare behaviors, such as "Attacking". In contrast, the 30-min sampling interval was able to detect differences among genotypes in high-occurrence behaviors, such as those associated with locomotory activity. Thus, from a practical viewpoint, when a broad characterization of chicken genotypes is required, the 30-min scan-sampling interval might be suggested as a good compromise between resources and results.

Figure 1

Bland–Altman plots of differences (as raw values) between data collected with 5- and 10-min sampling intervals in the mean proportion of animals engaging in each behavior per scan. The solid line represents the bias (− 0.0003), while the dashed lines represent the limits of agreement (− 1.836 to 1.835).

Figure 2

Percent of animals performing the different behaviors (% of all visible animals ± standard errors) per scan under the different sampling intervals. The behaviors are categorized according to their frequency of occurrence (low, medium, and high). A logarithmic scale was used for the y-axis to facilitate the visualization of data with very different frequencies.

Figure 3

Mean error rates (with error bars displaying 95% confidence intervals (CI)) for the behavioral recordings with 15- and 30-min sampling intervals. Data from the 10-min sampling interval were treated as the reference values. ns = no significant difference between 15- and 30-min sampling intervals.

Figure 4

Bland–Altman plots of differences (as raw values) between data collected with 10- and 15-min sampling intervals (Panel a) and data collected with 10-min and 30-min sampling intervals (Panel b) in the mean proportion of animals engaging in each behavior per scan. The solid line represents the biases, while the dashed lines represent the limits of agreement.

Figure 5

Bland–Altman plots of the percent differences between 10- and 15-min (Panel a) and between 10- and 30-min (Panel b) intervals. The solid line represents the bias, while the dashed lines represent the limits of agreement.

Figure 6

Odds ratios for the genotype effect (a, b, c and d) under the three sampling methods on high-, medium- and low-occurrence behaviors (Walking, Dust bathing and Attacking, respectively). ***p < 0.001, **p < 0.01, and *p < 0.05 versus 10-min interval data on each behavior.

Figure 7

Experimental area (Panel a) and timeline (Panel b) of the trial.

Figure 8

Graphical representation of the sampling intervals applied in the study (10, 15, and 30 min). Each scan was involved the same length of observation (10 s).