Presence/absence data are usually expressed as percentage or probability of sites being occupied. Thus, the data range from 0 (no site occupied) to 1 (all sites occupied). There are two main approaches that can be recommended for such data, depending on detection probability (see assumptions): logit link regression models or non-parametric regression models.
Assumptions. Detection probability either must be estimated and used to calculate the true percentage of occupied sites or must remain constant. Frequently, detection probability will differ among observers or sites (e.g. habitats). If it differs but remains constant in time, one still can analyse the data provided that sites can be grouped so that detection probability within groups of sites remains (approximately) constant. Then each group of sites can be analysed separately and the results can be integrated. Similarly, an analysis can be made separately for each observer. The assumption of a constant detection probability may also be violated because detection probability declines in many cases at low density (e.g. Henke 1998, Rodda et al. 2005). In this case, non-parametric regression models may be used for data analysis.
Analysis. Presence/absence data are first converted into the percentage of occupied sites (p). Then the data are analysed with a regression model with a logit link function of the form:
with t being time (e.g. year) and bi being the regression coefficients. The transformation of the percentage of occupied sites on the left hand site of the equation is called a logit or logistic transformation.
- Henke SE (1998): The effect of multiple search items and item abundance on the efficiency of human searchers. J. Herpetol. 32: 112-115.
- Rodda GH, Campbell EW, Fritts TH, Clark CS (2005): The predictive power of visual searching. Herpetol. Rev. 36: 259-264.
- Sokal RR, Rohlf FJ (1981): Biometry. Freeman, New York.
EuMon core team; July 2011