The out-groups should be closely related to the ingroup, but not a part of it. In your case, I would look at some large scale phylogenies of termites, to get an idea which groups are closely related to your families, and then select a few taxa as outgroups based on availability

This will depend on how genetically diverse thees families are.

Testing for situational saturation can be a way to asses if the sequences are suitable for phylogentic inference.

In theory the outgroup is the rest of life. In practice a smaller sample is used. For the two families you are examining you need to define your ingroup. When you say that you are looking at their relationships you seem to indicate the possibility of these families being most closely related to each other or to varous other families. So the first question is what is the larger group to which these families belong that appears to be very well established. For example is there a resolved node such as a 'superfamily' or other entity?

The most immediate outgroup would be the sister group of that 'superfamily'. But the scope of the outgroup needs to be carefully considered. For example, humans are well established as belonging to a large bodied hominoid clade (humans, chimps, gorillas, orangutans) and so if one were to examine their relationships the very least is to use the most immediate outgroup - lesser apes. But lesser apes are a small and structurally quite uniform group which may give misleading results. The next sister group, Old World Monkeys, gives a much larger comparative sampling and leads to different results.

But even with selection of outgroups there is the problem of outgroup sampling. In molecular approaches the sampling is lazy - usually with the claim or assumption that just one species is sufficient to root the ingroup tree. Dubious in my opinion.

Thank you all for the information.