Out of 1000 adults, you'd expect 125 to be women 15-30. That gives a margin of error for that subgroup of 9% (at 95% confidence level). They quoted 40%, so it's

Out of 1000 adults, you'd expect 125 to be women 15-30. That gives a margin of error for that subgroup of 9% (at 95% confidence level). They quoted 40%, so it's 95% likely to be between 31% and 49%. The size of the population doesn't affect the accuracy of a survey, only the size of the survey.