Altitudinal effects on water fluxes of deer mice: A physiological application of structural equation modeling with latent variables

An animal's physiological capacities, the environment it inhabits, and their interaction determine the animal's water flux (turnover rate). the proximate physiological determinants of water flux (e.g., urinary concentrating ability evaporative water loss, dietary water content, and oxidative water yield) are well understood. Excluding effects of diet type (e.g., herbivory vs. granivory) and free water availability (e.g., xeric us. mesic habitats), ecological and environmental effects on water fluxes are less well known. We used labeled water to study the effects of large altitudinal differences (similar to 3,000 m) on water fluxes of free-living deer mice. In absolute terms, high-altitude mice had higher water fluxes and field metabolic rates but were smaller, than low-altitude mice. A multiple regression with altitude coded as a dummy variable indicated that after adjusting for mass and field metabolic rate, altitude had no effect on water flux. However, using mass and field metabolic rate as independent variables in multiple regression is inappropriate because these variables clearly are not measured without error (an assumption of multiple regression). Consequently, we used structural equation modeling with latent variables to analyze the data more rigorously. We assessed the sensitivity of our conclusions to the reliability of field metabolic rate estimates by assuming that the error variance infield metabolic rate was 0%, 10%, 20%, or 40% of its observed variance. The direct effect of latent mass on water flux was sensitive to assumptions about the field metabolic rate error variance, but the effects of altitude were not. For all four error variance assumptions the structural equation models indicated that altitude had significant indirect and total effects but no direct effect on water flux. Hence, altitude had substancial overall effects on water fluxes, bur these effects are likely attributable to altitudinal differences in the thermal environment, not to changes in the partial pressure of oxygen. Physiologists commonly measure variables with error and frequently they may be able To obtain multiple measures or indicators of those variables. Consequently, we suggest that structural equation modeling, particularly with the use of latent variables, holds considerable promise as an analytical tool for physiological ecologists.