Question: Is there ever a situation where a fixed-effect model is preferable to a random-effects model? That is, if in a meta-analysis one study has an exceptionally large sample, a fixed-effect model will weight the large sample very heavily, under the assumption that a larger sample is more likely to approximate the "actual" effect size of the relationship we are interested in. If there is no evidence that the study with the largest sample also used the most reliable and valid assessment of the variables of interest, why should that one study be weighted disproportionately compared to other studies with smaller samples? Given this information, shouldn't researchers always use random-effects models for meta-analyses? Why?