First, the sample was assigned the exposure value that was observed at the 10th percentile and predicted values were obtained, then the sample was reassigned the value at the 50th percentile, and so on. This produced estimates of the probability of pregnancy at each exposure level that were based on the estimated associations between exposure and pregnancy in the original regression model and that accounted for standing on covariates.42 We did this for each of 5 age groups. As Fig 1 shows, higher levels of exposure were associated with a higher predicted probability of pregnancy, with 2 to 3 times the risk of pregnancy for youths exposed to high versus low levels of televised sexual content