I would accept this argument if the sample was unbiased but noisy. In this case it's extremely biased but (potentially) low in noise.

If people from Uganda aren't part of the target audience of this site, we won't get Ugandan user agents even if they happen to be a fair chunk of web users worldwide (certainly more than in my small but high-tech country.)