Population data are often noisy because they are composed of multiple signals. These signals may originate from biological, environmental, and sampling processes. A promising method to isolate these signals, when multivariate population time series are available, is a factor analysis. A few types of factor analyses are available; they include maximum autocorrelation factor analysis (MAFA), generalized autocorrelation factor analysis (GAFA), principal component factor analysis (PCFA), and maximum likelihood factor analysis (MLFA). However, the performance of these factor analyses to isolate underlying signals from population data has not been compared. I evaluated the performance of these statistical methods by applying them to simulated data sets and used these methods to investigate salmon abundance time series. Results/Conclusions
PCFA with a varimax rotation performed the best to isolate underlying signals, except when the signals were autocorrelated, MAFA performed better. MLFA performed slightly worse than PCFA; this probably results from restrictive model assumptions such as normality and independence among specific factors. PCFA, MAFA, and GAFA were applied to salmon escapement data. The results from the analysis using the three techniques suggest that the fluctuation in salmon population is composed of the signals from both river and ocean environments. This supports a previous finding with maximum autocorrelation factor analysis alone.