The Full Story
DATA
Data were of different types and of different scales (Table 1 and 2) as they were collected using different techniques. Behaviour characteristic data were ranked based on a predetermined list and scale, as were the reactivity score data. The vocalization data were the count of how many times the animal vocalized during the experiment and the eye area temperature and the heart rate data were collected using a FLIR thermal camera and a Polar H10 sensor, respectively.
Table 1. First 6 rows of raw data table, including header.
Qualitative Behaviour Assessment Data
Other Data
Table 2. Data types and units by variable.
Assessment of Variables
Data were discrete variables and not normally distributed, except 'Heart rate' and 'Eye area temperature' (Figure 4). Qualitative behaviour variables and 'vocalizations' were zero-inflated (Figure 5) and reactivity scores were categorical variables. As such, effective transformations were limited and not applied due to the robustness of the discriminant analysis.
Figure 4. Other measured response variables
Figure 5. Qualitative behaviour assessment variables
Data Transformation
As part of the analysis, data were scaled and normalized using
"cor = T, xtfrm = kendall"
in the discriminant analysis function in R, which is more appropriate for discrete variables than the default. Additionally, the scaling was done manually in R using the "scale(center = TRUE, scale = TRUE)" function to be able to graph the effect of the transformation on each variable (Figure 6 & 7). Although, the discriminant analysis is robust against non-normal variables not meeting the assumptions and additional transformations are less effective on zero-inflated variables.
Figure 6. Boxplot of distribution of all variables before transfromation
Figure 7. Boxplot of distribution of all variables after transformation by scaling and centering each