Meta-Analysis using RVE (robumeta) I'm runnig a meta-analysis, correcting for clustering of standard errors with the robumeta command. R gives the following warning: 

If df < 4, do not trust the results

In some cases my df is in fact <4. What does this tell me then? Does it simply mean that the number of included studies is likely too small (often aroung 5-6, but with more than 10 single effect sizes)?
Is there anything else I could do for these cases?
 A: The OP did not come back to answer her own question as suggested so here is the relevant quote from the originators:

Aside from issues of power, this can be particularly problematic when the degrees of freedom fall below 4 for t tests, where the t-distribution approximation no longer holds. In these cases, two approaches are recommended. First, if the degrees of freedom are much smaller than the number of studies, the analyst should carefully examine the covariate values, paying attention to leverage points and imbalances. The suggested approach here is to follow the rules of thumb and guidelines for dealing with “unusual” data (e.g., outliers, leverage, influence points) in regression. For example, if most of the covariate values are between 10 and 20, with one “outlying” value of 100, a strategy may be to remove or winsorize this extreme value; doing so will improve not only the degrees of freedom, but may also help with the interpretability of the findings, since this observation could be exerting large influence on the coefficient estimate as well. Second, if the degrees of freedom are very small, a lower p value should be used; for example, if p < 0.05 is used as a threshold elsewhere, for these cases p < 0.01 should be used instead (since the type I error is typically higher than stated in these cases).

This is from Tanner-Smith and colleagues article entitled "Handling Complex Meta-analytic Data Structures Using Robust Variance Estimates: a Tutorial in R" available here
