I am conducting an analysis of a survey about the quality of healthcare received by futre mothers in a maternity clinic. My variable is the number of times (observed frequencies) that future mothers have provided a compliment about the maternity services. I am using the Chi Square goodness of fit test to test the association between number of compliments and the age groups (there are 5). There is however a big problem. I have noticed some of my groups are very large (e.g. women in reproductive age) and others are much less represented (for example, women in 40+ or 18- years). When I create my expected frequencies, Chi square theory says that I should divide the total from the observed by however many subgroups I have, and that these should be my expected frequencies data. However, does this not oversee the problem that my sample subgroups are of different sizes? Should I adjust the expected frequency sizes relative to the size of each specific age group as opposed to spread the observed total equally amongst the groups?