State abortion legislation as a public good - before and after Roe v. Wade.

Conway, Karen Smith ; Butler, Michael R.

I. INTRODUCTION

Prior to the 1973 Roe v. Wade ruling, each state could choose whether to restrict the availability of abortions just as it could choose to regulate the availability or total amount of any publicly provided good. In this paper we extend the theory of public goods and collective decision-making to abortion legislation. Abortion legislation is treated as a publicly provided good that results from the collective decision-making of the electorate. The median voter theorem then provides a convenient framework in which to identify the determinants of abortion legislation. Using state-level data prior to Roe v. Wade, we estimate the public demand for abortion legislation and predict the likely outcome for each state if Roe v. Wade is overturned. Our predictions should remain relevant even if it is not overturned, inasmuch as the Supreme Court's decision in Webster v. Reproductive Services (1989) upheld states' rights to regulate or restrict the availability of abortion.

Identifying and estimating the public demand for abortion legislation, however, entails two complications not usually found in the typical public demand/median voter model pioneered by Bergstrom and Goodman |1973~. First and most serious, there are two kinds of demand for abortion legislation, which we refer to as external and private demand.(1) The external demand for abortion legislation stems from the external costs or benefits conferred on the voter by the availability of abortion (or the lack thereof) and is reflected in the preferences of all voters. Private demand derives from the direct effect the availability of abortion may have on an individual who may some day choose to have one; by definition, then, only women of childbearing age have a private demand for abortion legislation. The public demand of the electorate is thus a combination of these two kinds of demand.

The second complication concerns the econometric specification of the public demand for abortion legislation. Specifically, there are only a finite number of possible choices for abortion legislation. We define three categories: (1) abortion is prohibited with very few exceptions, (2) abortion is significantly restricted, and (3) abortion is relatively unrestricted.(2) This categorical variable, Y, suggests an underlying variable, revealed public sentiment towards abortion, Y*, and is therefore most appropriately specified as an ordered-response model. Using state-level data prior to 1973, we estimate an ordered-response model of the public demand for abortion legislation. Our empirical results identify the determinants of the public demand for abortion legislation, provide insight into the relative importance of external versus private demand, and allow us to predict the likelihood that abortion will be restricted or prohibited in any given state if Roe v. Wade is overturned.

It is illuminating to compare our methodology (and our ensuing empirical results) to that of Medoff |1989~ and subsequent research by Chressanthis, Gilbert and Grimes |1991~, and Gohmann and Ohsfeldt |1990a; 1990b~. Medoff |1989~ also estimates the public demand for abortion legislation as a function of the characteristics of the electorate; however, his model assumes that abortion legislation is the result of competing constituencies, such as women in white collar professions, certain religious groups or nonwhites. Our specification also includes these population characteristics (and others), but it derives directly from a theoretical model in which abortion legislation is a public good with a public demand consisting of external and private components. We differ on our empirical approach also. Medoff |1989~ uses each state's U.S. Senators' votes on the Hatch/Eagleton amendment as an indicator of the sentiment towards abortion and probable legislation if Roe v. Wade is overturned. He then uses the empirical estimates to predict (within sample) what would happen if Roe v. Wade is indeed overturned. In contrast, we use information about how each state behaved prior to Roe v. Wade to forecast how it would behave if that rulings' restrictions were again removed, taking into account that the preferences of the state may have changed in the interim. We argue that examining the actual legislation that resuited when states were completely free to prohibit or restrict abortion availability should be a better measure of public demand than Senate votes on an amendment that returns states to this situation by overturning Roe v. Wade.

Chressanthis et al. |1991~, and Gohmann and Ohsfeldt |1990a; 1990b~ build on Medoff in a different way by investigating the roles that legislator idealogy and shirking(3), in addition to constituent preferences, play in determining how Senators voted on the amendment. While our analysis does not explicitly consider legislative shirking, we argue that such shirking should be less important in an unrestricted climate since (1) the political cost of shirking will likely be higher inasmuch as voters would now perceive the legislator's power to directly determine the legal status of abortion, and (2) the legal status of abortion will ultimately be determined by state governments(4), who are (arguably) more accountable to their constituents. As a result we contend that the electorate will ultimately determine the legal status of abortion and that state law prior to Roe v. Wade, as opposed to the votes of U.S. Senators, is a superior measure of the public demand for abortion legislation.

II. A THEORETICAL MODEL

Formulating the Voter's Demand

A theoretical model of the public demand for abortion legislation must somehow account for the fact that voters may be affected both indirectly (through the abortions performed on others) and directly (the decision to have an abortion oneself) by the availability of abortions. All voters can be indirectly affected by the availability of abortions, and this external effect may be either positive or negative. For instance, parents of a teenage daughter may receive external benefits from knowing that abortions are readily available. Conversely, that same legislation may impose external costs on the voter if abortion conflicts with his or her religious or philosophical beliefs. Such external benefits or costs form the voter's external demand for the availability of abortion, which can be written as

|Mathematical Expression Omitted~

where e denotes the external demand, i = 1,...,|N.sub.j~ voters, j=1,...,M states, |X.sup.e~ is a 1 x k vector of taste variables influencing the external demand, and ||Beta~.sup.e~ are the parameters to be estimated.

For those voters who could have an abortion themselves, there is a direct effect of abortion legislation also. For instance, a woman of child-bearing age should not only consider the effects of the abortions performed on others (the external demand), but the effect that the availability of abortion could have on her own decisions. The latter effect forms the private demand for abortion availability. Medoff |1988~ estimates the demand for abortions as a function of price, income, and a number of taste variables. The private demand for abortion availability may be similar, but not identical, to the private demand for abortions, as the former relates to the potential of having a private demand for abortion. In particular, the private demand for abortion legislation does not necessarily have to be positive if voters find the ability to have an abortion oneself especially troubling. We can write the private demand for abortion availability as

|Mathematical Expression Omitted~

where p now denotes the private demand for abortion legislation, and |X.sup.p~ and ||Beta~.sup.p~ refer to the variables and parameters of the private demand function. The public demand for abortion legislation by voter i in state j, then, is the sum of the external and private demands, or

|Mathematical Expression Omitted~

where |Theta~ equals one if the voter could potentially have an abortion and equals zero otherwise. The model written in (3) is similar to that specified by Wyckoff |1984~ who estimates the "social" and "private" demands for education. Households with children in school possess both demands for education, whereas households without children in school express only a social demand. We posit that the demand for abortion legislation may exhibit the same type of dichotomy between voters who could have an abortion and those who could not.

Applying the Median Voter Theorem

According to the median voter theorem, if all preferences for abortion legislation are single-peaked (which seems reasonable), then the outcome of majority voting should reflect the preferences of the median voter. Applying the median voter theorem, then, suggests that the abortion legislation in any given state (prior to Roe v. Wade which prohibited certain outcomes), |Q*.sub.j~, reflects the public demand of the median voter, or

|Mathematical Expression Omitted~

where |Mathematical Expression Omitted~ and |Mathematical Expression Omitted~ are the characteristics of the median voter that determine his or her external and private demand, respectively. Prob(||Theta~.sub.mj~ = 1) is the probability that the median voter has a private demand, and is approximated by (|n.sub.j~/|N.sub.j~), where |N.sub.j~ is the total voting population and |n.sub.j~ is the population of voters who have a private demand (women of child-bearing age). Extending the typical public good/median voter model to abortion legislation and taking account of the two types of demands leads to an abortion availability demand equation that is a function of the characteristics of the median voter and of the probability that he or she also has a private demand for abortion availability.

While (|n.sub.j~/|N.sub.j~) is the logical and perhaps the only practical choice as an approximation for Prob(||Theta~.sub.mj~ = 1), its validity will depend upon the distributions of |Mathematical Expression Omitted~, |Mathematical Expression Omitted~, and ||Theta~.sub.ij~. In particular, it may tend to overestimate the probability that the median voter has a private demand, as illustrated by two realistic scenarios. Assume for simplicity that ||Theta~.sub.ij~ is distributed independently of the external demand and that the mean of the private demands is positive, or |Mathematical Expression Omitted~. Given these assumptions, (|n.sub.j~/|N.sub.j~) will tend to overestimate Prob(||Theta~.sub.mj~ = 1) if (1) the private and external demands are positively correlated, or (2) the distribution of external demands has a greater density in the middle of the distribution (e.g., a normal as opposed to a uniform distribution). In the first scenario, adding the private demand flattens the distribution of total demands and pushes voters with private demands to one or both (if private demand can also be negative) tails of the distribution. In the second, adding the private demand stretches the distribution of total demands to the right (more positive) and moves more voters with private demands out of the middle of the distribution than it will move into it. This disproportionate movement out of the middle is caused by the higher density of external demands around the center of the distribution. In either scenario, the probability that the median voter will have a private demand is less than the simple probability that any one voter will have one, or (|n.sub.j~/|N.sub.j~).

Although admittedly imperfect, we believe (|n.sub.j~/|N.sub.j~) to be the only objective measure of Prob(||Theta~.sub.mj~ = 1) available. And, given one or both of the above scenarios, we can be fairly confident of the direction of its potential bias. Because it will tend to overstate the probability that the median voter has a private demand, it will place too much weight on the private demand component and therefore bias the estimates of ||Beta~.sup.p~ towards zero. Thus, the factors influencing private demand may be even more important in determining state abortion legislation than our results suggest.

Econometric Specification

The model derived above yields a demand relationship for Q*, which can be interpreted as the revealed public sentiment towards the availability of abortion. First, we rewrite this relationship as a stochastic relationship,

|Mathematical Expression Omitted~

where |Epsilon~ is an error term capturing the random factors that influence both the demand of the median voter and the political outcome in each state. However, we do not observe Q*; instead, all we observe is the resulting abortion legislation, Q, which is a discrete random variable that is related to Q* in the following manner:

(6) abortion prohibited:

Q = 0 if Q* |is less than~ |Q.sub.a~

abortion restricted:

Q = 1 if |Q.sub.a~ |is less than~ Q* |is less than~ |Q.sub.b~

abortion unrestricted:

Q = 2 if |Q.sub.b~ |is less than~ Q*

The model described in equations (5) and (6) is thus an ordered response model that may be estimated using either ordered probit or ordered logit, depending on the assumed distribution of |Epsilon~.(5) In estimating the model, one cannot identify both the constant and the two thresholds, |Q.sub.a~ and |Q.sub.b~, so |Q.sub.a~ is typically normalized to zero. As a result, |Q.sub.b~ should be greater than zero in order to be consistent with the ordered response framework.

III. EMPIRICAL ANALYSIS

Description of the Data

We use the abortion legislation in each state prior to Roe v. Wade as our indicator of public sentiment towards abortion. States are classified into three categories depending on whether abortion was (1) prohibited except to save the life of the mother or in the case of forcible rape, (2) restricted, where specific conditions, such as preserving the "health" of the mother, must be met, and (3) unrestricted, where no legal restrictions are placed on abortions performed prior to the viability of the fetus.(6) Due to limitations posed by the available census data, we must use mean values for the total population aged 18 and over to approximate the characteristics of the median voter that influence the demand for abortion legislation. Clearly, this approximation is only valid if the voting population is representative of the total population, and if the median voter is indeed the individual with the mean characteristics. We also control for the fact that certain subgroups of voters may have different external demands by including the proportion of the population that belong to the subgroup as an approximation of the probability that the median voter will belong to that subgroup. For the private demand equation, we use the mean characteristics of the population of women of child-bearing and voting age, which we define to be between 18 and 44 years of age.(7)

Table I lists the variables included in the external and private components of the demand for abortion legislation in our broadest specification. |X.sup.e~ includes income, percentage with a high school education, percent married, doctors per 100,000 people, percent Catholic or Southern Baptist (both of which are religious organizations strongly opposed to abortion), population density, percent nonwhite, percentage of female-headed households in poverty, percent voting Democratic in the last gubernatorial election, and the proportion of TABULAR DATA OMITTED the population over 18 that consists of women aged 18 to 44.(8) Income, education, marital status, race and religious affiliation are included as factors that may be correlated with a person's sentiment towards the availability of abortion. Doctors, as a group, may also have different preferences towards abortion and may have a financial interest in the legal status of abortion. In addition, doctors per 100,000 may be an indicator of the availability of medical care, including abortions. Population density may capture both the extent to which voters are affected by the actions of others and the degree of urbanization in a given state. Percentage of female-headed households in poverty may influence the external demand as a proxy for the costs of the Aid to Families with Dependent Children (AFDC) program. For instance, abortion may reduce state expenditures for AFDC if it reduces the number of women and children eligible for the program. Finally, percent voting Democratic may capture any voter idealogies or preferences not explained by the other variables.

The variables in the private demand equation are similar to those used by Medoff |1988~. Mean income of women aged 18 to 44 may affect the private demand for abortion legislation both as a proxy for the ability to pay for an abortion (Medoff's 1988 argument) and for the opportunity cost of the time spent being pregnant with and then bearing an unplanned child. The percentage of women aged 18 to 44 having graduated from high school may lessen the private demand by increasing knowledge of contraceptives, as in Medoff |1988, 357~, or increase demand as yet another proxy for the opportunity cost of the time spent being pregnant.(9) Marital status has an impact on the potential private demand as the costs of bearing an unplanned child are likely higher for unmarried women. Percent nonwhite is included as another taste variable that may affect the demand for abortion, as in Medoff |1989~, and percentage of female-headed households in poverty again captures the effect of AFDC and its impact on a woman's private demand for abortion legislation. Finally, percentage of women 18 to 44 participating in the labor force reflects the higher opportunity cost of unexpected pregnancy, regardless of marital status, of women who are working for pay.

In order to predict the legal status of abortion in each state should Roe v. Wade be overturned, we must collect this information for the most current year possible. Unfortunately, for many variables 1980 is the most recent year available. In these cases, we predicted the current value by using a combination of national and statelevel growth trends of similar variables for which there was information available. The data sources for the variables in both years, as well as our method of constructing the current values are discussed in the appendix, and the means and standard deviations are listed in Table II. For completeness, we also report the number of states in which each variable increased between 1970 and the present.

Results of Estimation

We estimate the model written in (5) and (6) using both ordered probit and ordered logit analysis. Because the signs and statistical significance of the coefficients are the same across techniques and ordered logit consistently outperforms the ordered probit in prediction (within sample), we report only the results for the ordered logit model in Table III.(10) The "broadest" model includes all of the variables mentioned above and produces no statistically significant coefficients, although the overall goodness of fit of the model is quite high, yielding a Chi-squared statistic that is significant at a 0.06 percent level. Because our regressors are highly collinear(11) and in order to preserve a reasonable number of degrees of freedom, we drop several variables from our "broadest" model to arrive at our "best" model. We drop income, education and marital status from the external demand because of the little explanatory power demonstrated by the variables in the broadest model and the variables' high collinearity with the corresponding variables in the private demand equation. From a theoretical viewpoint, we argue that these variables should be more important in the private demand for abortion legislation. We also drop education from the private demand equation, arguing that most of the effects of education should be captured by the income and labor force participation variables.(12)

TABULAR DATA OMITTED

The results from our "best" model differ from the "broadest" model in the expected manner and are listed in the second column. First, several of the variables that are almost statistically different from zero in the "broadest" model become significant. Specifically, the number of doctors per 100,000 people and the percentage of the population that is nonwhite have positive effects on the external demand for abortion legislation (leading to fewer restrictions), and the percentage of the population belonging to either the Catholic or Southern Baptist church and the poverty rate of female-headed households have negative impacts. Income and the poverty rate of female-headed households have positive influences on the private demand for abortion legislation, whereas the percent of the female population that is nonwhite has a negative impact. Finally, the proportion of women of childbearing age, which belongs in both demand components, is significantly negative and, as expected, the estimate of the upper threshold (|Q.sub.b~ in equation (6)) is significantly positive.

One variable of interest that is not statistically significant is percentage voting Democratic, which we included in an attempt to capture any voter idealogies not captured by the other taste variables. We therefore find its statistical insignificance to be a reassuring result. The overall goodness of fit is diminished very little in the "best" over the "broadest" model and the omitted variables are not statistically significant (based on a likelihood ratio test) at a level as low as even 50 percent. The model correctly predicts the legal status of abortion in 37 out of 51 "states,"(13) and has a Chi-squared statistic that is significant at the 0.02 percent level. Because of its relative parsimony and ability to predict, we choose this specification as our "best" model.

Also of interest is the relative importance of the external and private demands in determining the legal status of abortion. Columns 3 and 4 in Table III report the TABULAR DATA OMITTED results of setting the private and the external demands, respectively, to zero. The proportion of the population that is women of childbearing and voting age is included in both demands, as the variable itself may affect the external demand for abortion legislation, as well as capture the constant in the private demand equation. In each case, we perform a likelihood ratio test, as suggested by Maddala |1983, 49~, to test for the significance of the omitted component and reject that either the external or the private demand is zero at a 1 percent significance level. Thus, both demand components are statistically important in determining state abortion legislation.

A final issue is whether the variables we include in |X.sup.p~ are really components of the private demand equation or if they are actually part of external demand. We can explore this concern by estimating the model,

|Mathematical Expression Omitted~

and testing whether ||Beta~.sup.p~ or |Mathematical Expression Omitted~ is equal to zero. On the basis of likelihood ratio tests, we can reject that ||Beta~.sup.p~ = 0 at a 10 percent level of significance, whereas we fail to reject that |Mathematical Expression Omitted~ at even a 50 percent level.(14) The data therefore supports our distinction between variables affecting the private as opposed to the external demand for abortion legislation and provides evidence that both demand components are statistically important in determining the legislative outcome.

Prediction for Each State

Using current values of the explanatory variables and the estimated parameters from our "best" model, we can predict the likely legal status of abortion if states can once again make unrestricted choices.(15) However, our approach has a potential for bias that depends on the specifics of any ruling that overturns Roe v. Wade. Specifically, if overturning Roe v. Wade will eliminate all constraints on state-level regulation of abortion, as is likely, then our approach will overstate the probability that abortion will be restricted. Such a bias occurs because the laws existing in 1970 may have evolved very gradually, without a clear "starting point." If Roe v. Wade is overturned, a clear "starting point" will exist for each state from which it must decide whether or not to deviate. In other words, there is a "cost" in moving from the status quo that is not adequately captured in our model. Thus, our predictions will, if anything, overestimate the restrictions placed on abortion if Roe v. Wade is overturned.

The upper portion of Table IV lists the legal status of abortion by state in 1970, with an asterisk(s) to denote if our model incorrectly predicted the status for a given state. The lower portion of Table IV lists the predicted legal status of abortion with an asterisk(s) to denote a change in legal status from 1970. The most striking trend in Table IV is how our model predicts that the country as a whole will place fewer restrictions on abortions than it did in 1970. Specifically, only twenty states will "prohibit" abortion (compared to thirty in 1970), and eight will place "no" restrictions on abortions (as compared to only four in 1970). Furthermore, if the legal status of a state is predicted to change, nineteen out of twenty-five times it is to a less restrictive status. The statistics listed in Table II, particularly column 3, provide a partial explanation for this phenomenon. From 1970 to the present, the number of doctors per 100,000 has increased in every state and a great deal on average, from 143 to 198. Also, the percent of the population that is nonwhite has increased and the percent belonging to either the Catholic or Southern Baptist church has decreased in over two-thirds of the states, and the real income of women aged 18 to 44 (multiplied by their proportion of the population) has increased in all of the states. However, looking at the other variables indicates conflicting influences on the total demand for abortion legislation.(16) Surprisingly, changes in the determinants of the external demand lead to fewer restrictions overall, whereas changes in the determinants of the private demand suggest greater restrictions.

In order to better determine the importance of the private demand component, we also made predictions using the parameter estimates from the external demand only model specification. These predictions are reported in Table V. Removing the private demand from our specification reduces the number of states that will prohibit abortion and increases the number that will either significantly restrict or place "no" restrictions on abortion. Thus, omitting the private demand component biases our predictions towards even less restrictive abortion legislation. Such a finding implies that the private demand component may not result in less restrictive abortion legislation, and indeed appears to polarize the states. This unintuitive result may be due to the fact that our private demand specification may also be capturing the different external demands of women of childbearing age, and the empirical results suggest that this difference varies widely. Such a conclusion appears consistent with casual observation.

TABULAR DATA OMITTED

TABULAR DATA OMITTED

For purposes of comparison, Table VI summarizes Medoff's |1989~ predictions and includes Newsweek's |1989~ predictions about what would happen in each state if Roe v. Wade is overturned. In general our predictions suggest that fewer states will prohibit or "abolish" abortion.(17) We conclude that our predictions differ because our model emphasizes the tastes and preferences of the general voting population, instead of political (and idealogical) factors such as the actions and opinions of current legislators (most of whom were elected at a time when abortion legislation was restricted by Roe v. Wade, and therefore may have faced a relatively low cost of shirking on abortion issues). Our results coincide with what most opinion polls suggest--that the electorate tends to be in favor of less restrictive abortion legislation than that espoused (at least until recently) by many of its political leaders.

IV. CONCLUSION

The legal status of abortion in the United States has increasingly come into question as a result of recent Supreme Court decisions (e.g., Webster vs. Reproductive Services) and changes in the make-up of the Court. It is not at all unlikely that the Court's landmark 1973 decision in Roe v. Wade striking down restrictive state abortion statutes could be overturned in the near future. If Roe v. Wade were to be overturned, each state would again be free to either prohibit or significantly restrict abortion. In this paper we have developed a theoretical model of the public demand for abortion legislation, estimated the model using the legal status of abortion in each state prior to Roe v. Wade, and then, using current data, predicted the legal status of abortion in each state should Roe v. Wade be overturned.

Our approach rests on the premise that in an unrestricted climate (such as in the absence of Roe v. Wade), public demand, rather than the expressed preferences of current elected officials, will ultimately determine the legal status of abortion in each state. In the current restrictive climate the political cost to legislators of shirking on abortion issues may be small, thereby possibly causing a divergence between the actions of government officials and the preferences of their constituents. In this context, it is not surprising that our predictions (which are based on the preferences of the electorate) differ from the predictions of studies that emphasize the behavior of current government officials. In the absence of Roe v. Wade, we argue that the cost of shirking will increase a great deal, and that the electorate will therefore ultimately determine the legal status of abortion. Already, with Roe v. Wade only under siege, recent elections have borne witness to several "conversions" by candidates to positions on abortion that more closely match those of their constituents. If such a trend continues, our predictions, which are based on the characteristics of the electorate and suggest a less restrictive stance toward abortion, should prove to be superior.

TABULAR DATA OMITTED

APPENDIX

State-level data was collected for the two populations being considered, (1) all persons of voting age, and (2) all females of both voting and childbearing age (defined to be between the ages of 18 and 44). Data for the population subsets is from the U.S. Bureau of the Census's Census of the Population; Detailed Characteristics. The 1970 Census Series was the source for the data prior to Roe v. Wade, and comparable data for 1980 was collected from the 1980 Census Series. Many variables were not available by age, by sex, and/or by state for any year more recent than 1980. Consequently, we forecast variables for 1988 using either state level growth trends (which are not reported by sex or age) or national level growth trends (which are reported by sex or age) calculated for the interim 1980-1988. Our choice of state or national growth trend hinged upon where the greatest variation appeared to be--between the sex and age groups or across states. Most of the data used in calculating growth rates is available in the Statistical Abstract series.

Variables referring to the state population as a whole (as opposed to a particular subpopulation) were more readily available. Population per square mile (population density) and the number of physicians per 100,000 residents are available for each state for both 1970 and 1988 in the Statistical Abstract series. Data for the number of Catholics and Southern Baptists are available by state for the years 1970 and 1980 and are published in Churches and Church Membership in the US: 1974, 1982, respectively. To obtain current state-level data, growth in national membership for the respective religious groups was used to project state figures for 1988. The gubernatorial election results data were obtained from the Congressional Quarterly's Guide to U.S. Elections |1985~, Almanac of American Politics: 1988 and the 1987 to 1989 volumes of the Congressional Quarterly Almanac. And, finally, the legal status of abortion in 1970 was found in From Crime to Choice. A more detailed discussion of the variables used and the sources in which they were located is available upon request.

1. Wyckoff |1984~ addresses a similar issue in estimating the public demand for education.

2. In this paper, we focus on the legal status of abortion in each state prior to Roe v. Wade, rather than the restrictions placed on the provision of abortions in publicly funded facilities. The latter restrictions are largely irrelevant prior to Roe v. Wade because any abortion (regardless of where it was performed) was significantly restricted or prohibited in the vast majority of states. Current legislation aimed at abortions in public facilities may simply be the states' responses to the legal constraint posed by Roe v. Wade.

3. See also Nelson and Silberberg |1987~.

4. Gohmann and Ohsfeldt |1990a~ make a similar point.

5. For further discussion, see Maddala |1983, ch. 2~.

6. The status of each state is listed in Table IV, where our predictions are also listed. The legal status is found in Davis |1985~.

7. This upper limit on the child-bearing age of women corresponds to Medoff |1988~. Also, the Vital Statistics of the United States |1990~ reports a rate of just .5 live births per 1000 women aged 45-49 years in 1970, as compared to a rate of 8.1 live births per 1000 women aged 40-44 years. Finally, the Current Population Reports of the Bureau of the Census (Fertility of American Women) reports fertility data for women aged 18-44.

8. We experimented with several other variable specifications, such as treating Southern Baptists and Catholics separately, and including the percent voting Republican, as well as the party affiliation of the current governor. In contrast to Medoff's |1989~ results, we did not find Southern Baptists alone to have a significant impact, and the percent Catholic alone typically was not significant either. Combining the two consistently produced a significant impact; this may be due to the concentration of Southern Baptists in the south and of Catholics in the northeast. As for the political variables, these seemed to have little impact regardless of the variable chosen. To make the model as parsimonious as possible, we used only percent voting Democratic. We believe this to be a superior variable because it indicates both the outcome of the election (and therefore the party of the governor) and also the degree of Democratic party affiliation. The complete set of results are available upon request from the authors.

9. Medoff |1988~ does not find education to be a significant determinant in an abortion demand equation. We also find it to be insignificant, and subsequently drop it from our "best" model.

10. Again, the complete results are available upon request.

11. Specifically, the condition index, which is defined as the ratio of the largest eigenvalue to the smallest eigenvalue of the data matrix, is 1076 which is indicative of extremely severe multicollinearity. For more discussion of the condition index, see Judge et al. |1985, 902~.

12. Auxilliary regressions on these variables as a function of the other K-1 independent variables produced R-squareds of .95, .92, .84, and .94, respectively, lending support to our decision to drop them from the model.

13. We include the District of Columbia in our study.

14. The full set of results is available upon request.

15. Clearly these predictions are valid only if the true parameters of the public demand equation have not changed since 1970. This is an admittedly stringent assumption, but a necessary one if we are to use our model to predict out of sample.

16. Specifically, changes in the percent voting Democratic and the percentage of female-headed households in poverty both acted to decrease the likelihood that abortion would be prohibited, whereas changes in population density, the proportion of voters with a private demand (|n.sub.j~/|N.sub.j~), (|n.sub.j~/|N.sub.j~) percent of women aged 18-44 who are married, (|n.sub.j~/|N.sub.j~) percent of female-headed households in poverty, (|n.sub.j~/|N.sub.j~) labor force participation rate of women aged 18-44, and (|n.sub.j~/|N.sub.j~) percent of women aged 18-44 who are nonwhite all acted to increase the probability that abortion would be prohibited.

17. We should mention here that our predictions are fairly robust to model specification. We constructed predictions using the "broadest" model specification and parameters estimated with the District of Columbia (a very influential outlier) omitted. Both sets of predictions are very similar to those projected by the "best" model, but suggest slightly less restrictive legislation. This, in conjunction with the bias caused by our model not capturing the "cost" of moving from the status quo that was discussed above, implies that the predictions reported in Table IV overestimate the restrictions placed on abortion.

DATA REFERENCES

Barone, Michael, and Grant Ujifusa. Almanac of American Politics: 1988. Washington: National Journal, Inc., 1987.

Congressional Quarterly Almanac: 1987. Washington: Congressional Quarterly, Inc., 1988.

Congressional Quarterly Almanac: 1988. Washington: Congressional Quarterly, Inc., 1989.

Congressional Quarterly Almanac: 1989. Washington: Congressional Quarterly, Inc., 1990.

Congressional Quarterly's Guide to U.S. Elections, 2nd ed. Washington: Congressional Quarterly, Inc., 1985.

Davis, Nanette J. From Crime to Choice. Westport, CT: Greenwood Press, 1985.

Jacquet, Constant H., Jr., ed. Yearbook of American and Canadian Churches 1989. Nashville: Abington Press, 1990.

Johnson, Douglas W., Paul R. Picard and Bernard Quinn. Churches and Church Membership in the United States: 1970. Washington: Glenmary Research Center, 1974.

Quinn, Bernard, Herman Anderson, Martin Bradley, Paul Goetting and Peggy Shriver. Churches and Church Membership in the United States: 1980. Washington: Glenmary Research Center, 1982.

U.S. Bureau of the Census. 1970 Census of the Population: Detailed Characteristics. Washington, D.C., 1973.

U.S.Bureau of the Census. 1980 Census of the Population: Detailed Characteristics. Washington, D.C., 1982.

U.S. Bureau of the Census. Current Population Series. Washington, D.C., 1988.

U.S. Bureau of the Census. Statistical Abstract of the United States: 1973, 94th ed. Washington, D.C., 1973.

U.S. Bureau of the Census. Statistical Abstract of the United States: 1980, 102nd ed. Washington, D.C., 1981.

U.S. Bureau of the Census. Statistical Abstract of the United States: 1989, 110th ed. Washington, D.C., 1989.

REFERENCES

Bergstrom, Theodore G., and Robert P. Goodman. "Private Demands for Public Goods." American Economic Review, June 1973, 280-96.

Chressanthis, George A., Kathie S. Gilbert, and Paul W. Grimes. "Idealogy, Constituent Interests, and Senatorial Voting: the Case of Abortion." Social Science Quarterly, September 1991, 588-600.

Davis, Nanette J. From Crime to Choice. Westport, Ct: Greenwood Press, 1985.

Gohmann, Stephan F., and Robert L. Ohsfeldt. "Predicting State Abortion Legislation from U.S. Senate Votes: The Effects of Apparent Idealogical Shirking." Policy Studies Review, Summer 1990a, 41-56.

-----. "U.S. Senate Voting on Abortion Legislation: A More Direct Test for Idealogical Shirking." University of Alabama at Birmingham, Working Paper Series #90-14, June 1990b.

Judge, George G., William E. Griffiths, R. Carter Hill, Helmut Lutkepohl, and Tsoung-Chao Lee. The Theory and Practice of Econometrics, 2nd ed. New York: John Wiley and Sons, 1985.

Maddala, G. S. Limited-dependent and Qualitative Variables in Econometrics. Cambridge: Cambridge University Press, 1983.

Medoff, Marshall H. "An Economic Analysis of the Demand for Abortions." Economic Inquiry, April 1988, 353-59.

-----. "Constituencies, Idealogy, and the Demand for Abortion Legislation." Public Choice, February 1989, 185-91.

Nelson, D., and Eugene Silberberg. "Idealogy and Legislator Shirking." Economic Inquiry, January 1987, 15-25.

Newsweek. "State by State: What Might Happen if the Supreme Court Overturns Roe v. Wade," 1 May 1989, p. 38.

U.S. Bureau of the Census. Vital Statistics of the United States. Hyattsville, M.D., 1990.

Wyckoff, James H. "The Nonexcludable Publicness of Primary and Secondary Public Education." Journal of Public Economics, August 1984, 331-51.

KAREN SMITH CONWAY is Assistant Professor, University of New Hampshire and MICHAEL R. BUTLER is Associate Professor, Texas Christian University. The authors thank Kelly Chaston for her superb research assistance, and Morris Coats, Adrienne McElwain Steiner and two anonymous referees for their comments.