Clarissa's Blog

When N. started getting very agitated about an article on teenage sexual habits in the Journal of Business and Economic Statistics, I wondered who its authors were. A very brief search revealed that the very first of the piece’s authors has published research “proving” that women are less competitive than men and that affirmative action is detrimental to “the blacks” who “cannot succeed” in good schools. After such a pedigree, what hot-button issue is left for such a scholar to explore? The answer is obvious: sex.

The article that annoyed N. so much attempts to prove that a reliable and constant access to contraception results in a higher rate of teenage pregnancies. In order to decrease the rate of teenage pregnancies, the article claims, it is crucial to avoid educating teenagers about contraception.

How would it possess one to defend abstinence education in 2012, you will ask. Well, here is the argument its authors make and also the reasons why said argument is deeply flawed.

Arcidiacono, Khwaja, Ouyang. “Habit Persistence and Teen Sex: Could Increased Access to Contraception Have Unintended Consequences for Teen Pregnancies?” Journal of Business and Economic Statistics, April 2012.

This paper attempts to model how teenage sexual behavior in the US would respond to changes in policies related to sex education and accessibility of contraceptives. The authors assume that teenage pregnancies are very costly to both the young parents and the society: having a child at the age of 14-19 tends to cripple the parents’ professional development and place a sizable burden on the government.

The decision making process of the teenage population is formalized in a nice discrete choice model where “choice” refers to whether a teen chooses to have sex and, if s/he does, what kind of contraception (condoms, pill, unprotected) s/he decides to use. In addition, the teen can choose between three levels of sexual activity (low, medium, high), which results in 10 combined choices. Authors propose a certain utility function that governs the choices made. E.g., for some teenagers, the “utility” of unprotected sex is greater than the cost of potential unwanted pregnancy, and so they engage in the former.

The utility function is based on a few assumptions that one may call “hard” and “soft”. The “hard” assumption must hold regardless of whether it is confirmed by the data. One such assumption is that teenagers adjust the level of their sexual activity depending on the accessibility of contraceptives. As the accessibility goes down, some teenagers supposedly reduce the frequency of sex or even switch to abstinence, and vice-versa.

Secondly, a teenager acts in a manner that maximizes the utility over the course of the entire teenage period, which is 4-5 years. The choice made today affects the entire future path. Since each year 10 options are available, in the beginning the teen is supposed to go over some 100,000 possible paths and pick the one the highest utility. In short, he/she is supposed to act like a very skilled chess player who thinks many steps ahead. Have you met many teens who analyze 100,000 possibilities before deciding to get laid?

The “soft” assumptions are soft in the sense that, if they are not confirmed by the real data, the estimate of the corresponding numerical parameter in the model will be insignificantly different from zero. The model assumes that sex might have a “fixed cost”, that is, there may be a “moral or psychological barrier that is crossed the first time one has sex” (this is a direct quote from the article, mind you). All other things (age, socioeconomic status, race, etc) being equal, a teenager who has had sex in the past is more likely to have it this year than a teenager who has abstained.

Another parameter in the model refers to the “transition cost of sex”. All other things being equal, a teenager who had sex last year is more likely to have it this year than a teenager who had no sex last year. Indeed, if a teen had sex last year, s/he is quite likely to have a sex partner who sticks around this year also.

The “fixed” and “transition” costs of sex amount to a phenomenon of “habit persistence” when, roughly speaking, today’s sex breeds more future sex. Once the teen becomes sexually active, it is hard to stop because s/he starts assigning more “utility” to sex. This implies that loss of virginity results in a sex exposure that is many times greater than the original single sex act. Therefore, if the goal is to reduce the sex exposure, it is a very good idea to increase the average age at which teenagers become sexually active.

The parameters of the model are estimated using a sample of about 4800 US teenagers born in 1980-1984, who were surveyed in 1997-2000. Such factors as race, age, school grades, household income, etc are taken into account. The dataset has no information about STDs, only pregnancies. Because the data do not include any measures of accessibility of contraceptives during that period, it is impossible to confirm or deny the first “hard” assumption, that teens reduce/increase their sexual activity in response to reduced/increased accessibility of contraceptives.

However, the “habit persistence” hypothesis is, indeed, confirmed by the data. The main point of the paper is to show that the following scenario is plausible:

1) Accessibility of contraceptives goes up in year one.

2) In response, teens increase their sexual activity. In particular, the average age of becoming sexually active goes down.

3) The sex exposure becomes especially high in the following years because of “habit persistence” phenomenon.

4) The teen pregnancy rate goes down in the first year, but, since contraceptives are imperfect, it goes up in the following years.

In other words, higher accessibility of contraceptives backfires, and the phenomenon of “habit persistence” is the main underlying cause.

However, the authors themselves admit that their way of modeling the “habit persistence” holds only as long as “persistence” cannot be explained by some person-specific factors that were not included in the dataset. Consider two teenagers, A and B, who appear identical according to the factors present in the dataset. Teen A has sex in year 1, 2, 3, and 4. Teen B has sex in year 4 only. The model explains this difference by saying that A got lucky in year 1, and having sex in the following years is largely attributed to the “habit persistence” phenomenon: today’s sex breeds more future sex. If blind luck had picked B in year 1, B’s sexual history would have been similar to that of A.

An alternative explanation of what happened is that A is nice, and B is nasty. In that case even if B had gotten tremendously lucky in year 1, B’s sexual history wouldn’t have been as intense as that of A. This undermines the entire “habit persistence” hypothesis together with the negative scenario above. There may be a few more similarly plausible explanations, but of course the corresponding explanatory factors were never measured. The authors try to model this “individual-level heterogeneity” using so-called mixture distributions, but there is a good chance this solution is inadequate.

Even if the entire model is fine, the negative scenario above is subject to a high contraception failure rate. In particular, the model suggests that making condoms 10% more effective at preventing pregnancy results in drops in pregnancy rates in all years. According to a source cited by the authors themselves (Trussel 2004), such increase is quite achievable simply by learning how to use condoms correctly.

In conclusion, this paper presents a nice example of a dynamic discrete choice model that is fitted on a real dataset. It ends up being practically useless and the authors seem to recognize that themselves. Despite that, they kick the dead horse once more by suggesting that an interesting topic for future research could be looking into how an easier access to contraceptives can facilitate promiscuity that leads to more STDs and unwanted pregnancies.

I don’t even want to imagine the mental equilibristics one needs to perform to arrive at a conclusion that contraception causes the STD rates to go up rather than down. Another brilliant idea of these authors is that an easy access to contraception makes people cheat on their partners. So if you are faithful to your partner, I’m guessing you must be unable to acquire any condoms. 

It is sad to see science at the beck and call of such blatant ideological manipulations.