TEEN BIRTHRATES IN CALIFORNIA: WHAT REALLY MATTERS?

By

Deborah Franklin

PPA 207 Quantitative Methods, Spring 1998
Professor Robert Wassmer
California State University, Sacramento

May 20, 1998

I. INTRODUCTION

This project is an examination of teen birthrates in 57 of the 58 counties in California and the factors that impact birthrates throughout the state. The purpose of the project is not only to identify the factors that are significant, but also to understand the reasons these factors are significant and their implications for public policy. Throughout the reading and research for this project, the underlying question was whether culture, economic status, educational status, and community and home environments are important factors in determining teen birthrates.

While teens have given birth throughout the history of America, it is not until the mid-1970s that teen births became the focus of public discourse. At that time, teen birthrates were actually declining. However, other demographic and social factors made teen births appear to be a growing social problem. The baby boomers entered their teen years, dramatically increasing the number of teen births. Teens were having sex at younger ages than in the past. Also, pregnant teens were not getting married as pregnant teens had done in the past. According to researcher Frank F. Furstenberg (1991), "It was only then that teenage childbearing — that is, early and unmarried parenthood — became a more prominent and socially disturbing trend."

In 1996, teenage parenthood in America was estimated to cost taxpayers $6.9 billion a year. Included in this estimate are the increased costs of welfare and food stamps benefits, medical care, incarceration, and foster care. When the related social costs are added in, the annual cost of teenage parenthood is astonishing. "The gross annual cost to society of adolescent childbearing and the entire web of social problems that confront adolescent moms and ultimately lead to the poorer and sometimes devastating outcomes for their kids is calculated to be $29 billion" (Maynard, 1996). With such a high price tag, it is no wonder teen birth is the focus of considerable attention.

There is also considerable controversy about the estimated costs of teen parenthood. Some researchers have found the estimates of the costs of teenage birth to be exaggerated. They argue the negative impacts of teen birth are overestimated because researchers fail to take into consideration the disadvantaged social and economic situation of teenage mothers before they become parents (Furstenburg, p. 128, 1991). However, even taking into account the preexisting social and economic deprivation of teen mothers, having a baby makes matters worse. According to the Alan Guttmacher Institute (p. 62, 1994):

Because most teenage mothers come from disadvantaged backgrounds, 28% are poor in later life. Theoretically, had they delayed their first birth to age 20 or older, an estimated 16% would be poor; in fact, however, only 7% of women who delay childbearing are poor later on, demonstrating that the initial disadvantage of teenage mothers is compounded by the early birth.

In response to both the social and individual costs of teen births, substantial amounts of public money are spent to prevent teenage pregnancy and birth. Governor Wilson and the California Legislature have considerably expanded pregnancy prevention programs. In 1996-97, $73 million in prevention programs were funded. The programs range from a media campaign, a male involvement campaign, and community grants for locally developed programs to funding for prosecution of statutory rape. Funding for prevention programs was continued in the 1997-98 budget (Sproul, 1997).

Teen birthrates are declining in California and in the rest of the nation. Nationally, birthrates decreased from 62 births per 1000 teenage women aged 15-19 in 1991 to 55 births per 1000 in 1996, a drop of 11.9%. California’s teen birthrate dropped 19%. The declining birthrates reflect decreased sexual activity among both female and male teens and increased use of contraceptives, particularly condoms (Sacramento Bee, 1998). California’s prevention programs include funding for birth control for teens and public service announcements stressing responsibility for sexually active teens. These aspects of our state’s prevention policy may be having an impact. Still, in 1996 there were 63,118 births to women aged 15-19 and 1,485 birth to girls under 15 in California. Because of the costs borne by society as a whole, and teenage mothers in particular, understanding the factors that significantly impact teen birthrates is an important step in the formulation of public policy to prevent teen births.

An unpublished paper, Teenage Pregnancy and Birth in California: Trends and Characteristics written by the Alan Guttmacher Institute (1994), provides an excellent overview of teen birthrates in California. In general, birthrates vary by race/ethnicity and poverty level. In 1990, the birthrate for Hispanic teens was 117 per 1000 women aged 15-19. The same year, the birthrate for African-American teens was 104 and 41 for whites/others. The average birthrate was 71 per 1000 California women aged 15-19. Differences exist within the Hispanic population as well. Mexican-born Mexican-American teenagers are less sexually active, yet more likely to become pregnant and to give birth than American-born Mexican-American teens. Due to low levels of contraceptive use, nearly three-fourths of sexually active Mexican-born teens become pregnant. American-born Mexican-American teens are as sexually active as non-Hispanic teens, however they are more likely to become pregnant and give birth than white teens. Teenage women who have grown up in poverty are more likely to become pregnant or give birth than more affluent teenagers.

The Guttmacher Institute also found birthrates vary by county. Nine counties have birthrates of 100 or more per 1000: Del Norte, Fresno, Kern, Kings, Lake, Madera, Merced, Tulare, and Yuba. Six of these high birthrate counties are Central Valley counties: Fresno, Kern, Kings, Madera, Merced, and Tulare. Five of the nine counties with high birthrates also have higher than average rates of teenage women living in poverty: Fresno, Tulare, Merced, Madera, and Yuba. These differences in county birthrates have policy implications and should be considered when policies and programs are developed to ameliorate the factors that impact teen birthrates. Other than the simple comparisons discussed, these findings made no attempt to control for the factors expected to influence teen birthrates. The regression analysis I conduct will control for those factors.

The remaining sections of this paper include a literature review, the model, the data, the regression analysis, and the conclusion. In the literature review section, four articles are summarized and their applications to this paper are discussed. The section on the model includes both the broad factors and the specific variables used in the model and their expected direction of effect. In the data section, there is a more specific discussion of the variables and three tables related to the variables. The regression analysis section includes regression results, including the results of tests performed and log specification regressions. Finally, in the conclusion, the elasticities, the confidence levels, and the magnitude of the coefficients are considered and some conclusions about the results of the regression are drawn.

II. LITERATURE REVIEW

Hoffman, Foster, and Furstenberg (1993) reexamined the costs of teenage motherhood. Recognizing that teen mothers often come from economically and socially disadvantaged backgrounds, they sought to separate the effects of teen parenthood from the effects of their background. Using data from the Panel Study of Income Dynamics (PSID), they compared teen mothers to their sisters. They found the negative effects of teenage birth were overstated as some researchers have recently claimed. However, they also found when family background is taken into account "the effects of teenage childbearing do not disappear, nor, indeed, are they particularly small, especially for economic status." In their analysis, they used a fixed-effect model. They acknowledged several problems with their fixed-effect model, including the question of whether women with sisters could be compared to women without sisters and the assumption that the home environment was the same for both sisters with each receiving the same resources. They also noted that if a teen mother had less potential than her sister, the model would overstate the effects of teen birth. While their fixed-effect model resulted in weaker estimates than the comparable cross-section model, the results indicated teen birth negatively affects economic and social well being. Teen mothers completed fewer years of education. Only 54% graduated from high school, but an estimated 71% would have graduated if they had delayed childbearing until they were 20 years old. The model also estimated teen birth almost doubles the probability the teen will be poor and decreases by more than half the probability she will be middle class. Interestingly, they also found family background had an effect on fertility; half of the sisters in the study shared the same fertility status. Applying this study to this research project, I sought variables to represent family background in my regression model, including family income and poverty status.

Leibowitz, Eisen, and Chow (1986) sought an understanding of how pregnant teens make decisions. Their premise was that economic factors influence teens’ choice to abort, to give birth without marrying, or to marry before giving birth. Their sample included enough Mexican-American teens to separately analyze their decision-making. The authors acknowledge problems with their economic model of decision-making and with their sample. Their economic model of decision-making requires the pregnant teen to think rationally about the expected costs versus the expected benefits of having a child. It assumes some knowledge of future earning potential and the actual costs of raising a child, as well as a rationality and maturity which may be lacking in pregnant teens. Also, their sample consists of pregnant teenagers who visited a free clinic in Ventura County, California. The clinic served women who "perceived their pregnancy as a problem" and was "the primary intake point for nearly all women who received abortions" in the county. The sample is very biased towards pregnant teens who are not planning to give birth. While the authors admit their sample is biased, they believe it is representative of teens who might be considering abortion. However, I believe this bias makes suspect their conclusions about why some teens choose to give birth. The sample is also fairly small, just 297 teens.

The model Leibowitz and her co-authors utilized was a conditional logit function. Their variables included proxies for the value of the teen’s time, giving more value to the time of pregnant teens enrolled in high school; a self-support dummy and an AFDC /Medicaid dummy; and dummies for Catholic and for Mexican-American ethnicity. The latter dummies assumed that Mexican-American teens would be more likely to give birth than choose abortion due to their cultural heritage.

Interesting findings in this study of Ventura County teens include age-differences in the teens’ decision making. Women aged 16 and 17 were more likely to give birth than 18 or 19 year olds, perhaps because the older group could more easily obtain abortions. Teens who reported higher grades in high school were more likely to choose abortion, and teens who had already dropped out of high school were more likely to give birth. Mexican-American teens were more likely to give birth and get married than other teens in the sample. However, the variable for Catholic was not significant and was dropped from the final regression model. Teens who were on AFDC or receiving Medicaid were more likely to give birth and not marry. Applying this study to my research paper, I included variables for Mexican-American cultural heritage, high school dropout rates, and female high school achievement in the regression model.

Lundberg and Plotnick (1995) considered the choice of pregnant teens to abort, marry and give birth, or give birth without marrying. They also considered the choice to become pregnant. They utilized a three-stage nested logit model and data from the National Longitudinal Study of Youth (NLSY). The three stages were pregnancy, abortion, and marriage. Their economic approach assumed teen’s choices would be influenced by both long-run costs and short-run costs. They considered racial differences, but included only non-Hispanic whites and non-Hispanic blacks. Because the authors looked at pregnancy rates as well as birthrates, they had to take into account abortion rates. I focused on birthrates because of the difficulty in finding reliable data on rates of pregnancy and abortion. Both of these rates are generally believed to be underreported. Lundberg and Plotnick found abortion was underreported by as much as 80% by black teens in the surveys. This underreporting could have led to biased estimates, but the authors do not believe that it did. However, it did make interpreting any results about black teens problematic and was cited as a possible reason for the inconclusive results. In my opinion, the data was questionable enough to discount any findings about black teens. White teens, on the other hand, reported about two-thirds of their abortions.

The authors focused on costs and incentives in their study, especially those created by state policy. State policies on funding and availability of contraceptive and abortion services were represented by dummy variables. Background variables included variables for the teens’ family structure, whether their mothers worked, their mothers’ education, the number of siblings, and religiosity. For whites, the higher the AFDC payment, the more likely the pregnant teen was to give birth without marrying. White teens were more likely to choose abortion if the state funded abortion services than if the state did not. The model for black teens yielded different results: no policy variable had a significant effect. States with more conservative policies regarding abortion and contraception and with lower AFDC benefits were associated with higher premarital birthrates. Also associated with higher birthrates were mother-only households. Teens with mothers who worked and had higher levels of education were less likely to become teenage mothers. In their conclusion, the authors noted mixed results. Economic incentives that affect the cost of becoming a teenage mother mattered to white teens, but did not seem to make a significant difference for black teens. Applying this analysis to my research paper, I included a variable for working mothers with children under 18 and for female-headed households.

King, Myers, and Byrne (1992) studied the demand for abortion to determine what economic factors influenced teens’ decisions to give birth or abort. They developed a model based on the economic theory of demand. Their data was from the National Longitudinal Study of Youths (NLSY). The variables in their probit function included predicted hourly wage rates, local unemployment rates, family income, poverty status, school enrollment status, age, ethnicity, and frequency of religious attendance. The authors raised concerns about the underreporting of abortion in the data that I believe make their results less reliable. Their results should be considered with this bias in mind.

Comparing teens who gave birth to teens who opted to abort, the study found teens who aborted lived in communities with lower unemployment rates, higher family incomes, and fewer families in poverty. Teens who aborted were also more likely to be enrolled in college, but equally likely to be enrolled in high school. Black teenagers gave birth more frequently than whites, while Hispanics had birth and abortion rates similar to whites. Teens with lower predicted wages and teens living in poverty were less likely to abort. The authors report high unemployment rates, low wages, and few job opportunities lower the opportunity costs of giving birth and are associated with a lower probability of abortion. Higher family income was associated with higher probability of abortion. White teens were most likely to abort, even when all other variables were held constant. The authors conclude economic factors are important in teens’ decision to abort or give birth. Teens in better economic circumstances are more likely to opt for abortion. Applying this study to my paper, I included variables for race/ethnicity, high school dropout rates, family income, and family poverty in my regression model.

III. MODEL

The dependent variable for my regression model is the birthrate per 1000 teenage women aged 15-19 for each of 57 counties in California in 1990. (Alpine County was dropped from the sample because data is frequently unavailable for this tiny county of less than 1,200 people in 1990.) Birthrates for teens 15-19 are appropriate for the purposes of this study because the vast majority of teen births occur in this age group. The broad factors expected to impact teen birthrates are culture, economic status, educational status, home environment, and community environment. Specifically, the model is shown as the following function:

Birthrate = ƒ(culture, economic status, educational status, home environment, community environment)

where,

culture = ƒ(Hispan, Black, Asian, Spanlan),
economic status = ƒ(Teenpov, Fampov, Mfaminc),
educational status = ƒ(Dropout, SATfem)
home environment = ƒ(FHH, Fcwork)
community environment = ƒ(Rural, Subco, Urbco).

Subco and Urbco are dummy variables with Rurco being the necessary ommitted variable. Predicting the expected direction of the effect for each broad factor is problematic. Within each of the broad categories are variables that are expected to affect the dependent variable in different directions. Culture variables are expected to have varying effects. Hispan and Spanlan are expected to be positively associated with birthrates because Hispanic teens are assumed to have cultural and religious values that discourage contraceptive use and support birth over abortion. Because California-specific studies have shown black teens in California are less likely to give birth than other ethnicities, Black is expected to negatively affect birthrates. The direction of the effect of the variable Asian is uncertain; anti-immigrant groups have claimed Asians give birth at higher rates than whites, but their claims are suspect.

The specific variables for economic status are also expected to have different affects. Birthrates are predicted to increase as teen poverty and family poverty increase, while birthrates are expected to decrease as median family income increases. I expect these outcomes because the opportunity cost of teen birth has been found to be a significant factor. Poorer teens have less to lose than teens in better economic circumstances and are more likely to give birth than abort.

The variables for educational status are predicted to affect the dependent variable in opposite directions. Birthrates are expected to be positively affected by the variable dropout because teenage mothers often have low educational expectations and poor school performance. SATfem is expected to be negatively related to birthrate. This negative relation is expected because this variable indicates outstanding achievement in high school; some studies indicate young women who perform well in school are less likely to give birth. Also, these teens could have expectations of higher future wages, which would increase the opportunity cost of giving birth.

For home environment the direction of effect for both specific variables is also contradictory. According to some researchers, growing up in a mother-only household (FHH) is a significant positive factor in teens’ decisions to give birth. Some of these same studies indicate a working mother (Fcwork) decreases the probability a teen will give birth and thus is expected to negatively affect birthrates.

Finally, direction of the effect the general factor of community environment is uncertain. Because some of the literature on California teen birthrates report that rural counties had higher birthrates while some national studies indicate the socio-economic problems of cities are likely to increase teen birthrates in urban areas, I am unsure whether Urbco will have a negative or positive affect on teen birthrates. Subco and Rural are also uncertain. The more suburban counties and counties with a higher percentage of their population living in rural areas could have lower birthrates if they have fewer of the socio-economic problems related to more urban counties.

IV. DATA

I selected the variables in the regression model based on my readings about teenage birthrates. Hispan, Black, Asian, Spanlan are all proxies for culture. Numerous studies reported variation in birthrates by race and ethnicity, indicating culture plays a role in determining birthrate. California-specific studies suggest that Hispanic women who were born in Mexico differ from their American-born counterparts, so I included Spanlan to capture that cultural difference.

The variables for the general factor of economic status are Teenpov, Fampov, and Mfaminc. All are appropriate and widely used measures of economic well being. Poverty is frequently reported to be positively related to teen birthrates. These variables seek to verify that conclusion.

I used Dropout and SATfem as proxies for educational status. These variables are data on the same age group as the dependent variable, Br1519. Failure at school is positively related to teen birthrates; Dropout captures that failure. High future aspirations are expected to decrease birthrates; female high school students scoring well on the SAT are likely to have such aspirations.

The variables I selected to proxy for home environment were FHH and Fcwork. Some studies have shown teens are more likely to give birth if they were raised in a mother-only household. FHH clearly represents such households. On the other hand, teenage women with working mothers are less likely to give birth. Fcwork is the percent of women in the work force who have children under 18 years of age, an appropriate proxy for working mothers.

For community environment, I selected variables that would capture some of the differences in urban, rural, and suburban counties. A few researchers had noted that California’s more rural counties had higher birthrates. More urban counties had lower birthrates. I included suburban counties because I thought their more stable, better off households could be related to lower birthrates.

The following three tables contain information about the variables in the regression model. Table 1 consists of a list of the variable names, a brief description, and the source for the variable. Table 2 reports the descriptive statistics for each variable. Table 3 is a correlation matrix for the independent variables.

Table 1: List of Variables with Sources

Table 2: Descriptive Statistics

Table 3: Correlation Matrix

1-HISPAN 2-BLACK 3-ASIAN 4-SPANLAN 5-TEENPOV

1-HISPAN 1.000000

2-BLACK .164419 1.000000

3-ASIAN .161377 .638826 1.000000

4-SPANLAN .988708 .122496 .126771 1.000000

5-TEENPOV .315847 -.102315 -.010822 .318752 1.000000

6-FAMPOV .398497 -.038923 -.121621 .423588 .790784

7-MFAMINC -.090633 .233104 .405762 -.118141 -.521172

8-DROPOUT .178639 .278979 .309272 .146020 .332110

9-SATFEM -.439075 .050467 .290119 -.426705 -.394404

10-FHH .616917 .483815 .380832 .591815 .451137

6-FAMPOV 7-MFAMINC 8-DROPOUT 9-SATFEM 10-FHH

6-FAMPOV 1.000000

7-MFAMINC -.679238 1.000000

8-DROPOUT .330512 -.131367 1.000000

9-SATFEM -.627851 .608950 -.075089 1.000000

10-FHH .593889 -.099698 .420588 -.380661 1.000000

1-HISPAN 2-BLACK 3-ASIAN 4-SPANLAN 5-TEENPOV

11-FCWORK -.258639 .064011 .249653 -.278287 -.507942

12-RURAL -.400774 -.527157 -.667878 -.352610 -.102348

13-SUBCO .241651 -.070585 .079235 .226356 .205765

14-URBCO .109684 .600896 .567942 .085779 -.142332

6-FAMPOV 7-MFAMINC 8-DROPOUT 9-SATFEM 10-FHH

11-FCWORK -.763050 .594179 -.209510 .598617 -.428092

12-RURAL .110265 -.488633 -.210090 -.191805 -.577169

13-SUBCO .009001 .100706 .052402 .034782 .275397

14-URBCO -.194853 .423513 .283740 .229821 .261714

11-FCWORK 12-RURAL 13-SUBCO 14-URBCO

11-FCWORK 1.000000

12-RURAL -.247371 1.000000

13-SUBCO .109156 -.470380 1.000000

14-URBCO .139666 -.508236 -.365703 1.000000

V. REGRESSION ANALYSIS

The regression results for six required regressions are reported in Table 4 . The first regression I ran (Best OLS in Table 4) was the model reported above. Before any corrections were made, 12 of the 14 coefficients were significant with greater than 85 percent confidence. Only Asian and FHH were not significant. The R² was 0.898.

Table 4: Regression Results

Statistical significance at greater than 99% in a two-tailed test denoted by ***, between 90 and 98% by **, between 80 and 89% by *.

Multicolinearity was found for two variables in the broad category of culture, Spanlan and Hispan. However, since both were significant I retained them. Interestingly, they have opposite directions of effect.

In earlier versions of the regression model, before I had found what I considered to be the best variables, multicolinearity had been found. The multicolinearity was most likely due to my scatter shot approach of trying any variable that looked promising. When I took a more parsimonious approach based on the theory I developed from my readings, the remaining variables were not multicolinear.

There are no misspecification or sample selection problems in the regression model. Each of the variables in the model was selected based on findings reported in others’ studies and there were no omitted variables. The variables were carefully chosen to proxy for the broad factors in the model and with the theory behind the model in mind. California has 58 counties, but the sample consists of only 57 of these counties. Alpine County was omitted because it is a county of less than 1200 people. With such a small population, it is more comparable to a town than a county. Therefore, it is appropriate to omit Alpine County from the sample.

The model does contain dummy variables for urban and suburban counties. I performed an F-test to determine whether the dummy variables should remain in the model. (See Best OLS, F-test in Table 4.)

The sum of the squared residuals (SSR) for the restricted regression was 4863.22 compared with the SSR of 3489.76 for the unrestricted regression. The value I calculated for F* was 8.27. Since F^C was only 3.23, I retained the dummy variables.

The LM test results indicated that some heteroskedasticity exists. The P value was 0.137, so there was an 86 percent probability that heteroskedasticity existed. I ran the regression on ET as an OLS with robust VC matrix (reported in Table 4 as Corrected for Heteroskedasticity). Running the regression this way resulted in all the variables being significant, 13 of the 14 at 90 percent or better confidence levels and the remaining variable at an 85 percent confidence level.

I was concerned some of the variables in the model were endogenous, especially Dropout. There is much discussion in the literature as to what impacts the other most, being a high school dropout or being a teen mother. I ran a 2SLS regression using Dropout as the dependent variable and replacing it with Public in the first regression. In the second regression I used Yfit to replace Dropout in the original model. The results of the 2SLS regression (reported on Table 4 under the heading Corrected for Endogeneity and Heteroskedasticity) verified my suspicions that endogeneity was a problem. Two variables, Hispan and Spanlan, that had been significant in every other regression were no longer significant. However, 10 variables were significant and the R² remained high.

I also ran both semi-log and log-linear regressions, both corrected for endogeneity and heteroskedasticity, and reported them in Table 4. In addition to the two dummy variables, Subco and Urbco, one other variable could not be changed to a log form. The unchangeable variable was Rural which has one zero value. Both of these regressions resulted in a lower R² and fewer significant variables.

VI. CONCLUSION

In this concluding section, I will discuss the regression model that is corrected for both endogeneity and heteroskedasticity, the fourth regression reported in Table 4. I believe it is the best regression of the several I ran for this paper. A careful reading of the literature indicated that birthrates and dropout rates could be endogenous. Correcting for endogeneity was necessary to obtain reliable results, as was correcting for heteroskedasticity. This fourth regression was chosen over both the semi-log and long-linear regressions because they had fewer significant variables than the other regressions.

The elasticities for the coefficients were calculated by dividing the mean of the independent variable by the mean of the dependent variable and multiplying the coefficient of the independent variable by the resulting quotient. The elasticities are the percent changes in the dependent variable, birthrate, as a result of a one-percent increase in the independent variable. The elasticities for all of the variables in the regression model are less than one. Only one variable, Fcwork, has an elasticity near one. The elasticities for all variables are reported below in Table 5.

Table 5: Elasticities

Hispan = 0.437; Teenpov = -0.516***; SATfem = -0.300***; Subco = -0.183***; Black = -0.036; Fampov = 0.420**; FHH = 0.068;

Urbco = -0.097**; Asian = -0.071**; Mfaminc = -0.171**; Fcwork = -0.990**; Spanlan = -0.384; Dropyfit = 0.560*; Rural = -0.432***

* Denotes coefficients that are significant with greater than 80 and less than 90 percent confidence
** Denotes coefficients that are significant with greater than 90 and less than 99 percent confidence
*** Denotes coefficients that are significant with greater than 99 percent confidence

Nine coefficients are significant at either 90 or 99 percent confidence levels; one co-efficient is significant at a lower confidence level. Importantly, Dropyfit was significant at an 89 percent confidence level even after the endogeneity correction. Its positive sign was expected. The five coefficients that are significant at between 90 and 99 percent confidence levels are Asian, Fampov, Mfaminc, Fcwork, and Urbco. The negative signs for Mfaminc, Fcwork, and Urbco were expected. I was uncertain about the sign for Asian. Some anti-immigration groups have claimed that immigrants from Asia have higher birthrates than non-Asians. I would not have been surprised if the sign had been positive, however it was negative. Holding all else constant, Asian teens were less likely to give birth than non-Asians. This result makes the anti-immigrant groups’ claim questionable. The only positive sign in these five coefficients was for Fampov, which I had expected.

The four coefficients that are significant at 99 percent or higher are Teenpov, SATfem, Rural, and Subco. The negative sign for Teenpov was completely unexpected. Everything that I read indicated that poor teens were more likely to give birth than abort. This was true for both national studies and the California studies. The results of my regression seem to indicate that for families in poverty the more teens in the family, the fewer births. Perhaps already disadvantaged families discourage the teens from giving birth when there are more teens in the household. I had expected SATfem to have a negative sign, which it did. I was uncertain about both Rural and Subco, which were both negatively related to teen birthrates. These results are not surprising in that both Rural and Subco should be indicative of more stable communities with fewer of the problems associated with urban areas, such as poverty, crime, and inadequate education. Interestingly, in urban and suburban counties, as proxied by the dummy variables Urbco and Subco, as the percent of rural population in the county increases, the teen birthrate decreases.

The absolute values of the elasticities of the significant regression coefficients range from 0.071 to 0.990. The magnitude of the impact of the coefficients with lower elasticities is less than the impact of coefficients with the higher elasticities. The magnitude of the impact of a variable is an important policy consideration. No matter how significant the coefficient, a small magnitude would suggest that policy changes affecting that variable would produce little effect. Assuming a 10 percent increase in the number of Asians in a county, the birthrate could be expected to decline by just 0. 71 percent. A 10 percent increase in median family income (from the median income of $34,876 to $38,363) results in a 1.71 percent decrease in the teen birthrate. If there was a 10 percent increase in the number of women in high school who score higher than average on the SAT, a 3.00 percent decline in birthrate could be expected.

The variable family poverty was both significant and positively related to teen birth rates. A 10 percent increase in the number of families living in poverty would result in a 4.2 percent increase in teen birthrates. Policy efforts to lower by 10 percent the number of families living below poverty could be expected to decrease births by 4.2 percent. Rural’s elasticity has approximately the same magnitude as family poverty’s. If the percentage of residents in a county living in rural areas increased by 10 percent, the birthrate would decrease by only 4.32 percent.

Three coefficients had elasticities of 0.50 or more. Changes in these variables could have sufficient magnitude to warrant the attention of policy makers. According to the regression results and the elasticity calculations, a 10 percent increase in teens living in poverty would decrease the birthrate by 5.16 percent. However, no reasonable public policy advocate would recommend that actions be taken to increase teen poverty rates. A better approach would be to decrease the number of high school dropouts. A 10 percent decrease in the percentage of teens that drop out of high school would decrease the teen birthrate by 5.60 percent. With an elasticity of nearly one, the variable Fcwork is worthy of serious consideration. A 10 percent increase in the number of working women with children under 18 would result in a 9.9 percent decrease in teen birthrates. Considering the mean birthrate is 66 births per 1000, this means approximately 6 less births.

The R² for this regression is .891. Almost 90 percent of the variation in birthrates in California’s counties is explained by the variables in the regression. On the other hand, just over 10 percent of the variation is still unaccounted for. The large R² does indicate the chosen variables are correlated to teen birthrates and suggests the reasoning behind their selection is sound.

The thesis question for this paper was whether cultural, economic status, educational status, and home and community environments are important factors in determining teen birthrates. The regression results have shown that they are. Economic status and educational status are particularly important. All of the variables in those broad categories were significant at an 89 percent or higher percent confidence level. Surprisingly, cultural variables were not significant in this regression. In the best OLS version of this regression, the variable Hispan was significant and positively related to birthrates, but when the regression was corrected for endogeneity that variable was no longer significant. Also in the earlier version, the coefficient for Spanish-speaking households was significant and negatively related to teen birthrates, but the coefficient was no longer significant after correcting for endogeneity. In the final version, only the Asian variable was significant and it was negatively related to birthrates. These results are contrary to other studies, which have repeatedly found Hispanic teens are more likely to give birth than non-Hispanic teens.

One of the home environment variables was significant, the percentage of working women with children under 18. I was particularly interested in this variable because some of the theory on which welfare reform is based states that having a mother in the work force is a positive role model for children, so reform policies should encourage (or force) women to work outside their homes. As the Fcwork variable has a negative relation to teen birthrates, the theory may be credible. Surprisingly, the variable FHH was far from significant, with a t-ratio of 0.245. This result was also contrary to others’ research that found mother-only households to be a factor in increased teen births.

The variables for community environment were all significant and negatively related to teen birth. These results are not as contradictory as they seem. The variable Rural represented the percent of a county’s population living in rural areas, according to census data. While this result was expected, it should be noted that some health policy researchers have noted that California’s highest birthrate counties are rural in nature and would find the results inconsistent with their work. While I consider the result for the variable Rural reliable, I do not consider the results on the other two variables to be reliable. The variables for suburban and urban counties are rather arbitrarily defined, making the regression results virtually meaningless. The counties are ranked by total population. The first 13 counties are defined as urban, the next 22 are defined as suburban, and the remaining 23 are defined as rural. There is no consideration of the population density, the number of large cities, or whether most residents live in rural, suburban, or rural areas. Without further study, it would be unwise to interpret these two measures.

While the magnitudes of the coefficient vary in their potential impact, the regression results do provide some direction for policy making to decrease the rate of teen births. Moving families out of poverty, keeping teens in high school, and helping female students succeed academically are all possible ways to lessen the rate of teen births. Bringing women with children into the workforce is another possible way to decrease teen birthrates; and since the Fcwork coefficient had the largest elasticity of any of the variables, it is worth exploring as a policy option. However, it is important to consider that Fcwork and Fampov are negatively correlated (-0.76on the correlation matrix). When more women with children are working, there are fewer families in poverty and fewer families in poverty lead to lower birthrates. Before implementing policies that encourage women with children to work, more study is warranted.

There is certainly potential here for a Masters’ Thesis, and there is room for improvement as well. Although it could be accounted for by data errors or individual qualities not discernable in aggregated county data, I am particularly curious about the 10 percent of variation that is not accounted for in the regression model. Given time, I would like to find variables for "taste" for birth, family violence, past sexual abuse, and accessibility of birth control and abortion services. I also wonder about the impacts of other community services and programs for teens. It would also be interesting to find out if the prevention programs the state has funded have had an impact. There are also a few puzzling results I would like to explore, especially the teen poverty variable with its negative coefficient value and whether my correction for endogeneity could have been better. In short, there is no shortage of ideas to improve this research paper and turn it into a thesis.

VII. BIBLIOGRAPHY

"Blacks lead way as U.S. teens show reduction in birthrate." The Sacramento Bee. 1 May 1998: Sec. B, p. 8.

Furstenberg, Jr., Frank F. "As the Pendulum Swings: Teenage Childbearing and Social Concerns"

Family Relations April 1991: 127-138.

Gerhenson, Harold, et. al. "The Presence of Coercive Sexual Experience Among Teenage Mothers" Journal of Interpersonal Violence June 1989: 204-219.

Hoffman, Saul D., E. Michael Foster, and Frank F. Furstenberg, Jr.. "Reevaluating the Costs of Teenage Childbearing" Demography February 1993: 1-13.

King, Randall H., Steven C. Myers, and Dennis M. Byrne. "The Demand for Abortion by Unmarried Teenagers" American Journal of Economics and Sociology April 1992: 223-235.

Leibowitz, Arlene, Marvin Eisen, and Winston Chow. "An Economic Model of Teenage Pregnancy Decision-Making," Demography February 1986: 67-77.

Levitt, Diane. Teen Families and Welfare Dependency in California: A Review of the Consequences of Teen Pregnancy, Welfare-Linked Intervention Programs, and Policy Options for Teen Pregnancy. Sacramento: California State Library Foundation, December 1994.

Lundberg, Shelly and Robert D. Plotnick. "Adolescent Premarital Childbearing: Do Economic Incentives Matter?" Journal of Labor Economics, 1995, vol. 13, no. 2: 177-200.

Marnard, Rebecca A., ed. Kids Having Kids: A Robin Hood Foundation Report on the Costs of Adolescent Childbearing. New York: The Robin Hood Foundation, 1996.

Moses, Anne. Teenage Pregnancy Prevention in California: 1995 Policy Roundtable Series Report. Sacramento: California State Library Foundation, July 1995.

Powell, M. Anne. Teen Pregnancy in California: Effective Prevention Strategies. Sacramento: California State Library Foundation, December 1994.

Powell, M. Anne. Teen Pregnancy Prevention in California: 1995 Policy Roundtable Series Report Addendum, Speaker Transcripts, and Selected Handouts. Sacramento: California State Library Foundation, July 1995.

Sproul, Kate. Issue Brief: California Strategies to Address Teenage Pregnancy. Sacramento: Senate Office of Research, April 1997.

The Alan Guttmacher Institute. Teenage Pregnancy and Birth in California Trends and Characteristics, an unpublished paper in Anne M. Powell. Teen Pregnancy in California: Effective Prevention Strategies. Sacramento: California State Library Foundation, December 1994.

The Alan Guttmacher Institute. Sex and America’s Teenagers. New York: The Alan Guttmacher Institute, 1994.

Return to Course Sylabi and Information Page