These are great plots. They look good and do not hide any information. They provide quick insights into the data, including its spread, any differences present, and the rough distributions. It is important to choose a graph that matches the statistical test conducted. For instance, if you show mean +/- SEM or SD, you should […]
This again is more for my use than anyone else’s. The focus of this post is doing ANOVAs (or linear models, after all everything is a linear model), in a nice readable and publishable way with transformations, post-hocs and graphs. Mostly for appearance reasons, this processes uses lots of packages. Packages should normally be avoided, […]
Mediation analysis is a simple idea that is easy to do, but an absolute minefield to interpret correctly. The principal is easy – perhaps one explanatory variable correlates with the dependent variable by mediating (effecting) an other explanatory variable. So say you find out that bullet wounds correlates with death and blood loss correlates with […]
Often you have a truck load of potential explanatory variables, that all might interact with each other, giving a multitude of potential ways the explanatory variables could relate to the dependent variable. You could painstakingly create every possible model or you could do a step-wise regression. Step-wise regression automatically adds and subtracts variables from your model and […]
Sometimes you know your data can’t follow a normal distribution even after transformation – such as count data or yes/no data. If this is the case, then generalized mixed modelling is the way to go. Basically it’s very similar to general linear modelling which assumes normality of the population distribution, but instead it assumes a […]
A Cook’s distance greater than 1 is a sign that this data point (or random factor) is having a disproportionate influence on your model and should be looked into. Note: I’m not normally a fan of removing data without a valid reason, for me, you need both a statistical and experimental reason for removal. #Loading […]
When the data is unbalanced or there are missing values, repeated measures ANOVAs fail to report unbiased results. Furthermore, all the data of one individual will be ignored if data from one time point is missing. Multi-level linear models gets around this by comparing models of the data, not the data itself, using log-likelihood/Chi squared distribution […]
A principal component analysis (PCA), is a way to take a large amount of data and plot it on two or three axes. It does this without knowing which groups the data belongs to, so if you perform a PCA, plot it, and the data clusters nicely into the experiment groups, you know there are […]