We first estimated what the risk disparities between LGBQ and heterosexual youth were before accounting for potentially invalid data. We then used a machine-learning algorithm to detect response patterns that suggested when youth were providing extreme or untruthful responses.
For example, we treated their responses with suspicion if they reported eating carrots four or more times every day and said they were impossibly tall. That means we gave less weight to their responses when we re-estimated all of the disparities. We then saw how the disparities changed after the potentially invalid responses were taken into account.—including steroids—injected drugs, cocaine, ecstasy and pain medication without a prescription were not as pronounced.
Yet, while some outcomes were susceptible to invalid data, others were not. For example, LGBQ boys and girls were about twice as likely to be bullied at school and two to three times as likely to consider suicide. This shows that not all outcomes are equally affected by invalid data.The Youth Risk Behavior Survey provides vital information on the health and behaviors of high school students.