Fixed or Random effect model for Meta analysis of parasite prevalence?

Question

I am trying to conduct a meta-analysis in R using the metafor package. My data set is sex-specific prevalence of parasitisms of 150 ish mammal species. The aim of my analysis is to see whether there is a bias or not in parasite prevalence (i.e. parasitism in mammals is male-biased, female-biased or balanced).

I would like to run three models, one for all parasites, one for blood parasites and one for gastrointestinal parasites.

Now, when it comes to choosing the meta-analysis model I am not sure what to choose. I know the assumptions on which both models are based but the the variability between the prevalence studies from my point of view is not much, the principal differences usually is just the sample size.

I read somewhere that I should take the heterogeneity as references... and here it is:

All parasites I^2 = 47.06%, Q(df = 146) = 243.5527, p-val < .0001 Blood I^2 = 50.81%, Q(df = 95) = 159.4225, p-val < .0001 Gastrointestinal I^2 =36.10%, Q(df = 50) = 78.4733, p-val = 0.0062

according to this should I use for the first two random effect model and fixed for the third??

I hope I explained myself clearly. Thanks in advance.

Answer 1

The two models have different underlying scientific assumptions so it is best to choose based on your science, not the observed heterogeneity (although people often do just that).

Put loosely the fixed effect model says "I know that there is indeed a single underlying effect and it is my job to estimate it.". The random effect model says "I know that there is no true single underlying effect but rather there is a distribution of effects and it is my job to estimate its mean and variance."

For more details see the article by Hedges and Vevea here