This part of the Appendix describes how we created the fertility-related behavior/outcome variables.
Sexual activity and contraception variables
For both males and females, we construct variables for sexual activity and (lack of) contraception use based primarily on three questions from the survey: (1) “About how many times have you had sexual intercourse since the last interview?”; (2) “Thinking about all the times that you have had sexual intercourse since the last interview, how many of those times did you or your sexual partner or partners use any method of birth control, including a condom?”; and (3) “About what percentage of time have you and your partner used a condom?” Other lead-in questions were needed to separate nonresponses from no activity. The 2002 and 2003 surveys required an additional variable. An indicator for whether the person used a condom in every sexual encounter superseded the question on how often the person used contraception.
The first two outcomes for whether had any sex and the number of times the person had sex since the last interview (or in the past 12 months for 1997) is based directly from the first question. The number of times the respondent had non-contraceptive sex since the last interview is calculated as the difference between the answers to the number of times the respondent had sex and the number of times the respondent had used birth control. We excluded from the contraception models a few cases in which there were more instances of “used birth control” than “had sex.” The variable for whether had any non-contraceptive sex is constructed from this variable. The variable for the percentage of time the respondent did not use birth control, for those who had any sex, is constructed by dividing the number of times the person had non-contraceptive sex by the number of times the respondent had sex. Sen (2002) uses both of these measures for contraceptive use in an analysis of how alcohol affects sexual behavior. These variables apply for both males and females and are available starting in the first round in 1997. The variable for the percent of time the respondent did not use a condom comes straight from the survey.
Questions about sexual behavior are often sensitive. To address expected nontrivial nonresponse, the NLSY-97 uses follow-up questions. “Refusals” and “I don’t knows” to the numeric question on the number of times the respondent had sex was followed by a bracketing question. While nonresponse to the initial question on the number of times the respondent had sex ranged from 6 to 12%, about 60 to 70% of the initial nonresponders, depending on the year, answered the bracketed question. For the contraception question, the follow-up question asks about the estimated percent of time the respondent used birth control. Exploiting these follow-up questions, we include in our analyses both the numerical answers and the follow-up answers. For the bracketed answers for the level of sexual activity, we use the mid-point of the range. For the highest bracketed category of “201 or more” for the amount of sexual activity a person engages in, we use a value of 250. For non-contraceptive sex, we impute the value for nonresponders based on the number of times they had sex and the bracketed value for the percent of time they used birth control.
Pregnancies, abortions, and births
For pregnancies, abortions, and births, we create different outcomes for males and females because of differences in the type of question asked of males and females and differences in their knowledge of whether a pregnancy or abortion occurred. It is important to identify the timing of the decisions and outcomes because we need to assign these decisions/outcomes to specific periods to match the economic data. Unfortunately, the lack of specificity and the lack of timing questions in some of these variables will cause some inaccuracies in our data, especially for males. The variables used for these outcomes have a surprisingly small percentage of nonresponses—typically less than 1 and 0% in some years for females and between 0 and 3% for males. First, we discuss the outcomes for females.
For pregnancies, our goal is to determine which intervals between interviews the respondent was pregnant. We assign the female respondent as having had a pregnancy begin in the period since the last interview if either: (1) she was pregnant at the time of the interview, (2) she had at least two pregnancies since the last interview, or (3) she had at least one pregnancy since the last interview and was not pregnant at the time of the last interview. With this measure, we may miss some pregnancies that are not known about at the time of an interview. For example, someone may have become pregnant 2 weeks before the 1998 interview and not known about it at the time of the interview. Therefore, in the 1999 interview, they did not have a pregnancy that began since the 1998 interview and may not count the one that started before the previous interview. These data start in the 1998 round.
For abortions, we use questions on the outcomes of each pregnancy the respondent had since the last interview or in the past year. We create a dichotomous indicator for whether the respondent had an abortion since the last interview. This variable is available starting in the 1998 round.
The birth data present a challenge for determining when the decisions are made. Having a child is typically the result of certain decisions on the intensity of preventative measures at the time of conception point and, possibly, a continuous set of decisions throughout the pregnancy to keep the child or abort, and all these decisions could depend on the economy. Thus, the decisions would carry over from one interval (between interviews) to the next interval if a woman were pregnant at the time of the interview.
Our strategy is to focus on the economic conditions at the time of conception. It would be more ideal to consider the time of conception through the first trimester for the period in which the decisions are made. However, we cannot identify the exact date of conception. Instead, we can determine whether the conception occurred since the last interview or before the last interview based on two questions: one on births from pregnancies since the last interview and one on births from previous pregnancies. If a female had a birth from a pregnancy since the last interview, we code her as having given birth. If she had a birth from a previous pregnancy, then we code her as having given birth (or technically, made the decision to give birth) in the previous period—i.e., in the period before the last interview. We use data starting in the 1998 round for this outcome because the 1997 question about pregnancies is not under a specified period.
For males, we use a different set of questions specific to males. For pregnancies, we use a question on whether the male caused a pregnancy since the last interview. Of course, it is possible that some males are not aware that they caused a pregnancy. “Refusals” and “Don’t Knows” account for roughly 1% of all responses—mostly, they were “Refusals.”
Subsequent questions asked about whether any of these pregnancies ended in an abortion and whether any ended in a live birth. From these questions, we create dichotomous indicators for having caused a pregnancy leading to an abortion
and having caused a pregnancy leading to a birth
. Nonresponses occurred in between 0 and 3% of the answers to these questions. These questions on pregnancies, abortions, and births for males are available starting in the 1999 survey round.
List of all explanatory variables used in all models
Sample sizes for Tables 3 and 4
Sample sizes for Tables 5 and 6
Comparison of coefficient estimates on the unemployment rate from models using clustering at the state-year level versus the individual level