We describe the derivation of periodicities from visual data for long-perio
d variables, and give expected errors for the derived amplitude, phase and
frequency. We extend this analysis by using a moving window to deduce varia
tion in phase and amplitude over time. The method is demonstrated on the Mi
ra-type variable T Cas, and shows evidence that the phase of the first harm
onic (period 222.4 d) varies relative to the fundamental (period 444.8 d) w
ith a period of approximately 3000 d. Our analysis parallels, but is comple
tely independent of, the work of Szatmary, Gal & Kiss and Bedding et al., w
ho both use an approach based on wavelet analysis.