Consider the following survey conducted by Google and Gallup about gender differences in computer science learning:
Among other findings, it says that in a survey (conducted in 2015-2016), 57% of 14-year-old girls and 43% of 16-year-old girls answered "yes" to the question of whether they have ever learned any CS. The study reports this as "Reported learning of CS peaks at age 14, then generally drops for both [boys and girls]" (boys also drop here, but less than girls)
However, the question asks if the respondent has EVER learned CS. Therefore, if this is true of someone at age 14, it will by definition also be true of that person when they reach 16. Thus, assuming respondents are answering truthfully, this difference must be a cohort effect rather than an effect of age. In other words, two years before this survey was taken (when all the current 16-year-olds were 14) the "percentage of 14-year-old girls who had ever learned CS" must have been no greater than 43%. Thus this percentage must have increased from 43% to 57% in the space of two years. If true it seems this would be an extremely positive development, yet it is not reported that way in the study.
Is this type of anomaly (where the fraction of respondents who answer "yes" to whether something has "ever" happened to them goes down in the same group over time) common in these types of surveys? Am I correct in the assessment that there must be a positive development here which was overlooked by the study authors? Or are there other explanations which I'm not thinking of?