I think the proportion bullied out of total (or percentage) in the sub-group would be a representative number. This pproprtion can be compared to prevailing proportions in the general population and of a gender and tests of significance (such as chi-square test) conducted.

The 1st issue is the representativeness of your sample. I suspect the sample is a convenience sample & thus it will not in general be representative of more general populations. Without a random selection process for obtaining the data points, statistical tests may not be meaningful. Each of these 3 methods imply a different model underlying the data, so the questions I usually ask researchers is: Which model do you think is correct? Or, is there an established theory that suggests one model is preferred?

If you are looking for a less scientific answer, I would suggest proportions because this is a mathematical concept that is understood by most lay persons. As long as one does not try to project this proportion outside of the observed sample, you are on sound scientific grounds.

You may include a descriptive statistics sections (relative frequencies, bar charts, crosstabs) in your analysis to report your findings and afterwards you should conduct hypotheses tests based on your research questions.

Hypothesis tests presume the sample is selected in an unbiased manner and have suspect results if the sample is biased. As I mentioned above, the selection criteria for generating the sample is the key to the usability of the sample in a statistical sense. Rather than advocating more sophisticated statistical techniques, I would suggest adding quality control measurements such as coverage rates, response rates, imputation rates, detected error rates, the number of outliers identified, etc. These indicators should be supplied for the full population & sub-populations estimated for. Such indicators will tell users the applicability of the data to the issues being discussed.