Emília P. Martins, Indiana University

1. Why won’t the newest version of COMPARE run on my computer?
2. How can I save my results to file?
3. How can I deal with missing data in COMPARE?
4. Why am I still getting a message that there is an error in my phylogeny? It works in xxxx!
5. Can I use COMPARE to analyze categorical data or to run an ANOVA?
6. There are no branch lengths on my phylogeny, what do I do?
7. How do I get a p-value for a correlation coefficient from COMPARE?

1. Why won’t the newest version of COMPARE run on my computer? Make sure you have a modern web browser (e.g., Internet Explorer or Netscape) and that you have recently downloaded and installed the Java Virtual Machine (http://www.java.com). COMPARE was written in JAVA and can be run directly from your browser without being downloaded or installed. It works well on PCs, Macs, and UNIX machines, with Internet Explorer and Netscape. You can also download an executable version or compile it yourself from the source code.
2. How can I save my results to file? Although, COMPARE does not write to or read from your hard drive, you can transfer your results with copy/paste (use the shortcut keys and right mouse button) to another file.
3. How can I deal with missing data in COMPARE? COMPARE assumes that each data set is complete. To handle missing data, create one or more reduced data sets and phylogenies. You can do this easily using the "Draw Phylogeny" button/window to delete taxa and save a new reduced tree.
4. Why am I still getting a message that there is an error in my phylogeny? It works in xxxx! If it's not a missing parentheses or comma, there is probably a polytomy in your phylogeny. Try using Phylodendron or some other program to draw the phylogeny to find it. For hard polytomies (representing rapid diversification rather than ignorance) insert very tiny branch lengths so that the phylogeny is actually bifurcating. For soft polytomies (representing ignorance), consider using the Generate Phylogenies page to create several possible phylogenies.
5. Can I use COMPARE to analyze categorical data or to run an ANOVA? Most of the analysis routines in COMPARE are set up as regressions. Hence, they work best when at least the Y variable is continuous. Categorical predictor (x) variables can be included, if you first code each variable as a set of dichotomous "dummy" variables. A one-way ANOVA with three levels is equivalent to a regression of y on two x variables (yes/no level 1, yes/no level 2). Look at a good regression textbook for further explanation.
6. There are no branch lengths on my phylogeny, what do I do? All phylogenetic analyses either use or assume branch lengths of one form or another. Most phylogenetic methods require that branch lengths be in units of the expected amount of phenotypic change for the character of interest, quantities that are almost never known. Lots of partial solutions have been proposed, including plots, statistical transformations, and randomization tests. My own preference is to use methods (e.g., PGLS, PMM) which estimate an extra parameter which automatically stretches and shrinks branch lengths to fit the data and phylogeny. In COMPARE, you can also generate 1000s of possible branch lengths for a known phylogeny, analyze your data on each and combine the results to calculate a confidence interval that takes your uncertainty into account.
7. How do I get a p-value for a correlation coefficient from COMPARE? Many correlation coefficients (e.g., for Felsenstein contrasts) can be tested against standard statistical tables. In other cases, it is difficult to know the distribution of correlation coefficients, making it even more difficult to calculate p-values. Instead, COMPARE reports regression slopes (b) and standard errors. If the error term in the regression model is independent, normally distributed, and homoscedastic, you can estimate a rough 95% confidence interval for a regression slope as +/- 1.96 standard errors. The correlation coefficient is directly related to the regression slopes. In a bivariate regression, if the regression slope differs significantly from zero, the correlation coefficient will too.