Thank you in advance for any advice on this ecological study of human use of rocky intertidal habitats!

I am examining the change in the number of human visitors (as well as those visitors then categorized into different behaviors, such as fishing, collecting, or tidepooling) to rocky intertidal habitats in southern California from a previous survey conducted in 1995 to one conducted in 2013. The goal is to determine whether management strategies to reduce use, particularly from illegal activities such as collecting, have been successful. The surveys were conducted on two day types (weekend and weekday) over three seasons (fall, winter, spring) in both 1995 and 2013. This was done at 8 sites total but I am interested in patterns within a site, therefore site is not a factor. I am most concerned with whether the number of visitors (or number of individuals engaged in one of the behavior categories) has changed between the two years and less concerned with whether there are differences in season or day type. The reason for distributing samples among seasons and day types is that there are established differences in use among the seasons and between weekdays and weekends and I wanted to ensure that these differences are incorporated in the model and equally weighed among years. In other words, spring tends to have less use due to the timing of low tides….if there were a high number of surveys in spring in 2013, results may be skewed towards a larger reduction in use that reality. Ideally, sampling would have occurred with equal numbers of surveys per year per season per day type. However, that is not the case as sample size varies from 2-6 surveys within a season and day type (e.g. 2 surveys for weekends in fall, 4 surveys for weekdays in spring, 6 surveys for weekdays in winter, etc…).

I am currently running a 3 factor ANOVA (year, season, day type) but I am not sure that is the appropriate approach. Furthermore, given our main focus on change over the two sampling years, our ability to simplify and state yearly differences are complicated in some cases by significant interactions. Is there a better approach for examination of the effect of year that is of concern? Is there a way to focus on the primary factor (year) but standardize the weight of the different seasons and day types? I have considered reducing the surveys so that there are equal numbers of surveys done across years X season X day type. In one case (ie. for one site), that would result in a reduced, standardized sample size of 2 surveys per year, per season, per day type which is not conducive to analyses.

Also troubling in the analyses are the high number of zeros, particularly for some sites where use is very low or instances where collecting did not occur in 2013 (even though collecting was high in 1995 suggesting management has worked). While the data appears to have equal variances, the data fails normality tests.

I would greatly appreciate any advice, statistics appears not to be my strong suit.