1

I conducted an experiment in an evolutionary program called AVIDA-ED contrasting the adaptive evolution in different world sizes (30x30,40x40,50x50).

I performed 10 repetitions of each world size and recorded 4 variables being: Organisms (# organisms in world),Viable (how many organisms are viable), Fitness (average fitness of all organisms, the focus variable as it represents the reproductive quality), and Energy (the rate of acquirement an organisms learns to express a particular energy).

My alternative hypothesis is that adaptive evolution will occur at a higher rate in a large world (50) than a small world (30). Providing Fitness, Viability, and Energy = adaptive evolution, would a comparison of data from each world size using t-tests in R be appropriate? If not, which statistical tests would be advisable for this set of data to compare Fitness and Viable values and to visualize the data? From looking at the data, it is already able to be assumed that the larger world allows higher rates of all variables but I'd like to be able to represent that statistically. I'm very new to R so sorry for the cluelessness.

Dataset: https://gist.github.com/anonymous/a549abfaf949c08172b8c03642c2776f

jdsurfer
  • 45
  • 4

1 Answers1

1

Usually the t-test assumptions include either the normal distribution of random variables or large sample size (> 30), or both at the same time in order to generate unbiased statistic.

Since you have just 10 observations per sample, the usage of the t-test is not so straightforward, however, if the desired effect size is small compared to observed statistic difference you can still solidly detect the difference. (Let's say you see t-score equal 10, you can be quite sure the difference is out there, but if it is just 2, you can decline the Null hypothesis with Type 1 error chance which will be higher than 0.05 (0.025)).

You could also check out the power of the test to be sure your statistic distributions are far from each other.

One more option is using member count as an independent variable in a linear regression model, and then you are about to study the z-score and p-value associated with the covariate.

One of the discussions that gives much more details can be read here: How to choose between t-test or non-parametric test e.g. Wilcoxon in small samples

Alexey Burnakov
  • 2,469
  • 11
  • 23