Linear regression is designed for response variables that are continuous, they can take on a large, if not infinite, number of values. Binary response variables, either/or can only take one one of two, coded as 0/1.
Logistic regression is designed for discrete variables. The most common type, and the one you are using is binomial. The R syntax to deal with this class of problem is
fit <- glm(choice ~ age, type = 'binomial')
You get somewhat different diagnostics. High p-values will tell you right away if there is a lack of association, but low p-values require goodness-of-fit analysis.