If you are doing PCR with the intention of sequencing the product and checking a short region of the PCR product for a mutation, is there a region where taq polymerase is more or less likely to make a mistake? For example would it be more likely taq makes a mistake near the primers or in the middle of the PCR product? Should I be using a high fidelity enzyme even if it is only a short (20-25 bp) piece that I am concerned with?

As a rule of thumb the error rate scales most strongly with the number of reactions (i.e. length and cycles). The lenght is not an issue here, so keeping the cycle number as low as possible can help. Think of it that way, during any single reaction there is a (mostly) stochastic chance of an error, and then there is the chance that this molecule is used as template (and hence increasing erroneous products within the population). Sequence also has an influence, though unless you have runs (i.e. a stretch of identical bases) it is most likely less of an issue. Of course, using high fidelity polymerases are always preferred.

Remember you do not amplify the 25 bp once, but many times and the error rate is a function of the total reactions occurring within the sample.

So in the PCR product is it more likely that a mutation will occur in the middle, near the primers or are they equally likely?

The position relative to the primers has a relatively low influence. Other factors are more important.

Not really answering your question, but if you're concerned with the errors Taq may introduce, use a higher fidelity enzyme like Accuzyme (Bioline).