For my final project I am analyzing a data set of my choice and using the statistical methods I learned in the course. However the project I chose requires me to do some research on statistical methods for time-dependent variables as we did not learn them in the course (the professor approved my project but said I needed to do this). I found a data set with the size of insect eggs for 6700 insect species that was collected over a couple hundred years. I am comparing the size of these egges between the 8 genuses over time (bigger smaller no change). I have done a little research and found methods like the Cox Proportional Hazard Model but not I am not sure if I should use this. Are there any other methods? Maybe a two-way ANOVA? (which I did learn in this course). Any input would be greatly appreciated.
What statistical method should I use to comapare variables over time?
- Thread starter zbenzerara
- Start date
- Tags statistical methods
Cox Proportional Hazard Model is designed for events in time, such as death after a period of time. If you have a finite number of groups spread out at time intervals, you might try a repeated measures ANOVA. If you think your professor would allow you to research alternative methods, you could try Shewhart control charts and control charts. These are used in industrial statistics to determine whether a process is changing over time or is stable and repeatable.
If you wanted to do something simple you could take the data (separately for each species and project forward several years) with exponential smoothing. You need at least 50 points of data.
All the time series methods I know which show how one variable influences another over time are way to complex for most courses unless you have a really high end course. There are corrections for SE that are fairly simple but the impact can change over time. This is an issue I think even with cox models (unless you assume time invariant effects).
All the time series methods I know which show how one variable influences another over time are way to complex for most courses unless you have a really high end course. There are corrections for SE that are fairly simple but the impact can change over time. This is an issue I think even with cox models (unless you assume time invariant effects).