This is a definition I got from the statistical dictionary:
"random effects model is a statistical model that assumes some of the features of the model are randomly chosen from a wider population. Two common assumptions are that subjects are a random selection of all patients with the target disease and that study centres are a random sample of all centres that treat the disease."
Now suppose I conducting an epidemiological study (effect of exposure to certain a particular chemical) in a country. I divide the country in say 5 regions and I take samples of subjects (exposed and unexposed) from each region: this constitutes 5 trials. Now I know that the subjects are random samples from larger populations, but the trials (5= 5 regions) cover all the country. So I have randomness in subjects but no randomness in trials. According to definition above random effect is more attractive when you have randomness in both subjects and trials. But I'm at a loss whether fixed or random effect is a better model to go for this meta-analysis.