You're Outta Here: Why Californians Supported Legislative Term Limits

By

Jean Kinney Hurst

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

May 26, 1999

I. Introduction

The passage of Proposition 140, the initiative limiting the terms of California’s state legislators, became a major turning point in California politics. This paper examines the potential factors involved in voters’ decision to support Proposition 140. While term limits are a popular political concept nationwide, relatively little research focuses on the California initiative and the factors involved in its support. This paper examines the characteristics of the political climate, the economic climate, and California’s 58 counties, as well as county demographic characteristics, to determine the sources of political support for term limits in California.

Proposition 140 experienced a turbulent ride from passage to implementation. After numerous judicial challenges, in March 1998, the United States Supreme Court affirmed the ruling by the Ninth Circuit Court of Appeals upholding term limits. The term limit law limits Assembly members to three terms of two years each and Senators to two terms of four years each. Additionally, legislators are subject to a lifetime ban for each office held and are no longer given pensions upon retirement. This new set of regulations has tousled the statewide political process, resulting in new rules for a very old game.

While term limits have delivered a need for a revised political strategy systemwide, they have also opened doors for local government elected officials to advance to the state legislature. (More than one half of the 33 freshmen elected in 1996 to the Assembly had some local government experience.) Apparently, to many voters, local experience holds statewide value. This result has the potential to fundamentally change the relationship between the state legislature and California’s counties. Historically, counties have shared an uneasy relationship with the state; battles over the return of property tax revenue, transportation funding, and welfare reform have been fought for years, and, for the most part, lost. However, with an increasing number of Assembly members and Senators coming from city councils, boards of education, and county boards of supervisors, counties have a unique opportunity to gain legislative support from their own alumni.

The state-county relationship and the means by which it can be repaired are subjects that interest me greatly. By looking into the causes for support for term limits, I believe counties may find a source of political assistance for their causes. Perhaps voters were seeking a means for government by "real citizens" or an increased respect for local control in the Capitol and supported term limits to accomplish those goals. Perhaps the influx of former local government elected officials into the legislature was only an unintended consequence of term limits support. In any case, I have no doubt that counties must pay close attention to the possible causes for voters’ choices as expressed via the initiative process. Proposition 140, the term limit law, is certainly one of the most notable votes in recent history and has markedly impacted the function of California’s political system.

The following pages will include a discussion of articles relevant to this study, a detailed description of the theoretical model, a report of the data, an evaluation of the regression analysis, and concluding comments. The literature review will summarize and apply four articles associated with term limits. The description of the model used for this project includes a description of the dependent variable and the broad factors expected to influence the dependent variable, as well as the specific variables used to proxy for the broad causal factors and their expected direction of effect. A description of the data will follow, including specific statistical data for each variable. An evaluation of the regression analysis will include the regression results and various specification choices. In the conclusion, the elasticities, statistical significance levels, and other important analyses will be used to determine the validity of the regression experiment.

II. Literature Review

The concept of term limits has come to the forefront of political debate only recently and relatively little research has been done focusing on causes of voter support. Certainly, even less academic study has focused on California, and I was unable to find any article that looked at the local government issues that I include in this study. However, I have reviewed four articles to gain a more thorough understanding of the causes of support for legislative term limits.

Todd Donovan and Joseph R. Snipp use post-election opinion data to determine the basis of public support for Proposition 140. Specifically, Donovan and Snipp were interested in demographic characteristics, partisanship, and the effects of campaign contact via television, radio, newspaper, or mass mailing. They hypothesize that those underrepresented in the legislature may support term limits as a means of gaining proportionate representation (p. 493). In California, specifically, the authors expect Hispanics and Asians to be among the groups supporting term limits. This hypothesis is tempered, however, as it requires that voters possess substantial information about the potential consequences of a ballot proposition. In addition, due to the long-term Democratic majority in the legislature, Republicans and conservatives may also view term limits as advantageous.

Donovan and Snipp collected opinion data and ran a logistic regression with three models. The first model is composed of demographic traits only; the second adds an indicator of conservative ideology; the third adds three indicators of campaign exposure. Their regression results showed that women and young voters tended to support Proposition 140. Still, the under representation hypothesis was not supported. Republican partisanship also showed a significant, positive effect on the dependent variable. This article certainly gave me numerous validations for including certain independent variables in my own regression model.

Priscilla L. Southwell discusses how cynicism and inefficacy plays a role in supporting Congressional term limits. She believes that those individuals who express a certain type of alienation or cynicism are more likely to favor term limits, a "throw the rascals out" response, while those who feel powerless or inefficacious are more likely to refrain from voting and remain relatively indifferent to the issue of term limits, a "throw in the towel" response (p. 741-742.) Ms. Southwell conducted a probit analysis using opinion poll data to test her hypotheses. She found that wealthier individuals, as well as those who identify with the Republican party, are more likely to favor term limits, while male individuals tend to oppose them. In accordance with her hypothesis, cynical individuals favor term limits while less efficacious individuals show no distinctive preference. This study was more difficult to apply to my own research, as it is difficult to determine a proxy for "cynicism" or "inefficacy" in the county-level data I chose to use; however, certain groups of people may be more or less cynical or inefficacious, thus the inclusion of seniors, bachelor’s degree or higher, racial characteristics, poverty, and various employment measures.

In a similar vein, Jeffrey A. Karp uses survey data from the 1992 American National Election Study and other statewide surveys to examine explanations for support of Congressional and legislative term limits. Specifically, Karp looks at cynicism, self-interest, and ideology as causes for support. The author notes some interesting facts, including that, between 1990 and 1994 voters in 21 states approved initiatives limiting the number of terms legislators may serve (p. 373). In fact, term limits initiatives passed in almost every state where they appeared on the ballot, most with little or no opposition. This widespread support indicates to me a considerable need for further study of the phenomenon of term limits.

Karp uses logistic regression to determine the probabilities for each independent variable. He finds that the overall fit of the model is poor, suggesting that, either alone or together, the variables cannot account for the total support that term limits has received. He initially finds no significant effects of partisanship, until political knowledge is considered; as political knowledge increases, Republicans become more supportive, while Democrats become less supportive. Cynicism shows the strongest and most consistent effect in the model. Karp concludes that individual support for term limitations can best be explained by cynicism and to some extent self-interest and not ideology or dissatisfaction with legislatures. Again, in using this article to assist in the specification of the model I use, I’ve included variables to represent partisanship and groups that may be more or less inclined to cynicism.

Using a more theoretical approach, Andrew R. Dick and John R. Lott, Jr. attempt to explain why rational voters simultaneously re-elect their incumbents and vote for term limits. Their explanation focuses on a free-riding problem, where even when all districts would benefit from replacing their political representatives, any individual district will find it costly to remove its own incumbent whose long tenure in office has made him skilled at directing government wealth transfers. So, term limits offer voters a means to coordinate their actions and overcome the free-riding problem (p. 1). This is especially the case in California, where, when voters passed Proposition 140, they also re-elected 96 percent of all incumbent state representatives and senators whose terms they sought to limit. Among these incumbents, 77 percent had served as long as or longer than the maximum term prescribed under their state’s term limit initiative.

While this article was not as directly related to my regression study, it provided some interesting insight into voters’ behavior that was not offered in other studies. Because voters understand that the greater the relative experience or tenure of their legislator, the larger the reduction in benefits voters sacrifice by removing him, term limits offers a way to lower the costs incurred by any district that removes its incumbent. This theory may play out in my regression study when considering the voting behavior of those districts represented by longtime Assembly Speaker Willie Brown and Senate President pro Tempore David A. Roberti.

III. Model

The dependent variable for the regression model is the percentage voting ‘yes’ on Proposition 140, the legislative term limit law, for each of California’s 58 counties. As I am searching for the characteristics that influence support of the initiative, this choice seems a logical one. I believe that the broad factors involved are the political climate, the economic climate, characteristics of the county, and other demographic characteristics.

The specific model is as follows:

Percentage Yes on 140 = f(political climate, economic climate, county, demography), where

• political climate = f(percentage registered Democrat, percentage registered Republican)
• economic climate = f(median household income, percentage in poverty, percentage homeowners, percentage unemployment, percentage employed in agriculture, percentage employed in manufacturing)
• county = f(fiscal health (revenues - expenditures), percentage revenue from property tax, region, charter or general law)
• demography = f(percentage Hispanic, percentage Black, percentage Asian, percentage school age children, percentage seniors, percentage with bachelor’s degree or higher)

Thus,

Percentage Yes on 140 = f(percentage registered Democrat, percentage registered Republican, median household income, percentage in poverty, percentage homeowners, percentage unemployment, percentage employed in agriculture, percentage employed in manufacturing, county fiscal health, percentage county revenue from property tax, region, charter or general law county, percentage Hispanic, percentage Black, percentage Asian, percentage school age children, percentage seniors, percentage with bachelor’s degree or higher)

Predicting the expected direction of effect for each broad cause is a difficult task, as each broad cause includes variables that have differing effects. However, research suggests that people who feel that they are underrepresented in the political process may be more likely to support term limits. In this model, then, I would expect the same result.

The two variables included in the political climate should have different effects. Historically, Democrats have been the majority party in California’s legislature. Thus, Republican registration should have a positive effect, as term limits may be seen as a means for gaining political power in the legislature. I am uncertain about the effect of Democratic registration, though, as those historically underrepresented in the political process (i.e. ethnic minorities) tend to align themselves with the Democratic party.

Variables that measure the economic climate may also have unlike effects, however, I am uncertain as to what those effects may be, as theory and research do not offer a clear expectation. They are included because the economy can often affect political ideology and play a deciding role in voting behavior.

County characteristics are included to determine whether local government issues play a role in determining support for term limits. If a county is in poor fiscal health, voters may seek "new blood" at the state level to assist in funding local services; conversely, if a county is doing well financially, voters may see no reason to change their state representation. Thus, I expect counties’ fiscal health to have a negative impact. I am uncertain, however, how the percentage revenue from property taxes may play into this model. The regional dummies were included to determine if, at the regional level, support for term limits vary. I believe that there will be a difference among regions, as history indicates that political differences among regions in California is commonplace, however, I am uncertain as to the direction of the effect. Counties established under general law may have a different political climate than those that have been established under charter, but, again, the expected direction of this effect is uncertain.

The demographic characteristics also provide differing effects. Research predicts that those who are underrepresented in the political process may be inclined to support term limits. Then, Blacks, Hispanics, and Asians should display a positive effect on the dependent variable. The remaining factors of percentage school age children, percentage seniors, and percentage with a Bachelor’s degree or higher are uncertain, as none of these characteristics directly coincide with a specific political ideology.

IV. Data

Variables were chosen to gain a broad perspective on the causes of support for term limits. The literature indicates that personal demographic data and political and economic data play roles in that support. I was particularly interested in the effects of the characteristics of the county and region, so those variables were included as well.

Democrat and Republican registration data were used to proxy for the political climate. According to various studies, political ideology tended to matter in voters’ support of term limits. As California’s legislature has been dominated by the Democratic party for many years, I expected that this study would produce a similar effect.

The economic climate can drive one’s political decisions. Thus, I included median household income, percentage in poverty, percentage homeowners, percentage unemployment, percentage employed in agriculture, and percentage employed in manufacturing to measure the economic climate of each county. Each is commonly used to determine economic well-being.

To characterize each county, I included a measure of the county’s fiscal health (determined by subtracting expenditures from revenues), the percentage of county revenue received from property taxes, region, and charter or general law county. These data were included to provide an opportunity to draw conclusions as to the influence of the plight of local governments in voters’ support for term limits.

Demographic characteristics were included in accordance with the literature. Since those who are underrepresented in the political process tend to support term limits, I included those racial categories that are historically underrepresented in the California legislature: Blacks, Hispanics, and Asians. Percentage school age children, percentage seniors, and percentage with a bachelor’s degree or higher were also included as these characteristics can influence political choices.

The following tables provide specific data on each variable used in the regression model. Table 1 lists the variable name, a short description, and its source. Table 2 reports the descriptive statistics for each variable. Finally, Table 3 is a correlation matrix for each variable.

Table 1: List of Variables and Sources

*Note: Los Angeles County is considered the sixth "region" and is not included in the data set.

Table 2: Descriptive Statistics

Table 3: Correlation Matrices

* denotes highly correlated variables (.800 or greater)

Table 3: Correlation Matrices, continued

* denotes highly correlated variables (.800 or greater)

V. Regression Analysis

Table 4 shows four regression results: the original OLS regression without specification corrections (without regional dummies), the OLS regression with the regional dummies, the log-log regression, and the log-log regression corrected for heteroskedasticity.

Table 4: Regression Results

* indicates 80-90 percent confidence levels
** indicates 90-99 percent confidence levels
*** indicates greater than 99 percent confidence levels

The standard OLS regression resulted in a single variable, Hispanic, as significant at the 91.6 confidence level and an R-squared of .653. Since I believed that the model should include some measure of regional variation, I included five regional dummy variables (Los Angeles County being the sixth and not included in the data set). I performed an F-test to determine if the variables had an effect on the model; the calculated F-statistic was 52.428 and was greater than the critical F-statistic of 9.55, at 35 degrees of freedom for the numerator and five degrees of freedom for the denominator, at one percent level of significance. Thus, I preserved the dummy variables in the model.

The second OLS regression, including the dummy variables, resulted in no significant variables, but an increase in the R-squared to .733. Since I inferred that the data set may be better suited to a curvilinear regression line, I conducted a log-log specification regression analysis. This method proved a much better fit to the data. Five variables were significant with greater than 80 percent confidence and the R-squared value jumped to .812.

Concern regarding the possibility of heteroskedasticity was confirmed when I ran the Park test. I captured the residuals of the previous regression equation, squared them, and used them as the dependent variable in a separate regression with the independent variable of Z, the proportionality factor. I used the percentage of the statewide population in each county as the Z. I found that this Z was statistically significant at the 98 percent confidence level, evidence that there are heteroskedastic patterns in the residuals with respect to Z. To remedy this problem, I used the Weighted Least Squares technique and divided the equation through by Z, the percentage of the statewide population in each county. This adjustment resulted in an increase in the R-squared to .960 and ten statistically significant variables at the 80 percent or greater confidence levels.

I checked for multicollinearity by running a bivariate correlation matrix. The matrix indicated that two sets of variables were correlated, poverty and unemployment, and Democratic and Republican registration. However, both poverty and unemployment were significant at a high confidence level, 99.9 and 90.8, respectively. Democratic registration became insignificant after correcting for heteroskedasticity, but Republican registration became significant at the 86 percent confidence level after the correction; however, both retained the expected signs and were repeatedly referred to in the literature as theoretically valid. I felt it appropriate to keep both of them.

VI. Conclusion

The remainder of this paper will refer to the final regression analysis, the log-log specification corrected for heteroskedasticity. Since using the log-log specification results in coefficients that can be interpreted as elasticities, please see Table 4 for the elasticities for each independent variable. These elasticities can be interpreted in terms of the dependent variable; when the independent variable changes by one percent, the dependent variable changes by the corresponding coefficient value. For example, when bachelor changes by one percent, the ‘yes’ vote on Proposition 140 increases by .213 percent.

Five variables are significant at the 90 to 99 percent confidence interval. Income, homeowner, unemployment, the Bay Area counties dummy, and Asian are all significant within this interval. Homeowner, unemployment, and Asian positively effected the dependent variable, while income and Bay area counties had a negative effect. Two variables are significant above the 99 percent confidence level: bachelor and poverty, with a positive and negative effect, respectively. The other three significant variables lie in an 80 to 90 percent confidence level; manufacturing employment, county revenue from property taxes, and Republican registration all display a positive effect on the dependent variable.

While I was initially uncertain as to its outcome, the positive effect of the percentage with bachelor’s degree or higher may indicate a higher level of voter education as to the term limit law or perhaps it is a result of increased cynicism, as suggested by the literature. Well-educated people may follow state politics closely and become increasingly frustrated with the results. The magnitude of the elasticity of the variable indicates that a 10 percent increase in the percentage of persons with bachelor’s degrees or higher would result in a 2.13 percent increase in support for the term limit law.

The negative effect of income level is puzzling, as some findings in the literature suggested that wealth conferred a more conservative ideology indicating a likeliness to support term limits. However, those with higher incomes may oppose term limits to preserve the status quo. In this case, a 10 percent increase in median household income would confer a 5.82 percent decrease in support for term limits.

Poverty is also strongly negative, indicating that the more people live in poverty, the more likely an opposition to term limits. This may be the result of partisan ideology; Democrats tend to champion the needs of the poor and vehemently opposed limiting legislators’ term limits. A 10 percent increase in percentage of persons living in poverty (which is an ethically inappropriate policy choice) would result in a 3.54 percent decrease in support for term limits. One wonders how opponents of Proposition 140 would have reacted to this finding.

While I assumed that the income variable and homeowner variable would tend to give similar results, homeowner had a positive effect on the dependent variable. This finding correlates with the literature that suggests that those with a more conservative ideology were more likely to support term limits; an argument can be made that homeowners may have a more conservative ideology due to home ownership and the financial responsibilities involved. Here, a 10 percent increase in the percentage of owner-occupied homes would result in a 3.25 percent increase in support for the term limits initiative.

The positive effect of unemployment makes theoretical sense; the greater the unemployment, the greater the incentive to "throw the rascals out" and elect new state legislators to assist in boosting the economy. A 10 percent increase in unemployment (another inappropriate policy concept) would result in a 2.23 percent increase in support for term limits.

The positive effect of manufacturing employment is not so clear, however, those working in the manufacturing industry may also have an incentive to move perceived "anti-business" Democrats out of office. Here, a 10 percent increase in manufacturing employment confers an increase in term limits support of .03943 percent. Incidentally, this is the smallest effect of the significant variables.

I was surprised to see that the percentage county revenue from property taxes has a positive effect; this may be due to the fact that those counties with a higher percentage had more resentment towards the state which holds property tax allocation authority. In this case, a 10 percent increase in percentage county revenue from property taxes (wouldn’t counties love to see that?) results in a 2.33 percent increase in support for Proposition 140.

The negative effect of the Bay Area counties dummy is likely due to the potential effect of term limits on the local Assemblyman, Speaker Willie Brown, certainly one of the most powerful people in the history of California politics. Here, if a county is in the Bay Area region its voters are .0883 percent less likely to support term limits. Again, a rather small effect.

Asians tended to support the term limit law, in accordance with the literature. In this case, a 10 percent increase in the Asian population results in a .0454 percent increase in support for term limits. Finally, another expected effect, Republican registration had a positive effect on support for term limits. Here, a 10 percent increase in Republican registration confers a 1.98 percent increase in term limits support. This result is the most likely to effect policy decisions, as increasing voter registration is relatively simple to achieve.

The R-squared for this regression is .960. The R-squared, or coefficient of determination, is a measure of the "goodness of fit" of the model. A value of R-squared close to one shows an excellent overall fit. In this case, the model describes 96 percent of the variation in percent support for Proposition 140 in California counties.

In evaluating my thesis question, I have determined that some causes of the ‘yes’ votes on Proposition 140 were found in the broad factors of political climate, economic climate, characteristics of the county, and demography. Economic factors are particularly important when considering voters’ choices, reminiscent of James Carville’s now-famous quote "It’s the economy, stupid!" Demographic factors were surprisingly not as important; the only significant variable being the percentage Asian. The characteristics of the county, too, were not as important as I had hoped; however, I believe that better variables exist for measuring this effect. Political factors displayed the predicted effects; however, I did think that Democratic registration might be a significant factor. I should have expected the actual result, however, as Democratic registration is high in California, and, since Proposition 140 passed with a majority vote, many Democratic voters must have supported the measure.

One notable inference is that many characteristics of the economy and demography can be related to political ideology. While the relationship may not always be clear, personal experiences and the state of the economy are unquestionably involved in political decision-making. This link should be more carefully reviewed in attempting a project such as this to ensure appropriate analysis of causal factors.

I am considering a project for my Master’s Thesis that would discuss the California legislature post-term limits and the resultant effects of Proposition 140 for local government, paying close attention to counties. With additional time and resources, I would enjoy an opportunity to improve this regression analysis, especially in the characteristics of county factor to determine if frustration with the state-county relationship played a greater role in the support of Proposition 140. I am also interested in the cynicism factor reported by much of the literature. It would certainly be an informative addition to this analysis, if I could determine the appropriate proxies. Regardless, there is much to be learned from California voters and the case of Proposition 140 is no different.

Bibliography

Dick, Andrew R. and John R. Lott, Jr. "Reconciling Voters’ Behavior with Legislative Term Limits" Journal of Public Economics January 1993: 1-14.

Donovan, Todd and Joseph R. Snipp. "Support for Legislative Term Limitations in California: Group Representation, Partisanship, and Campaign Information" The Journal of Politics May 1994: 492-501.

Karp, Jeffrey A. "Explaining Public Support for Legislative Term Limits" Public Opinion Quarterly Fall 1995: 373-392.

Southwell, Priscilla L. "Throwing the Rascals Out" versus "Throwing In the Towel": Alienation, Support for Term Limits, and Congressional Voting Behavior" Social Science Quarterly December 1995: 741-748.

Studenmund, A.H. Using Econometrics: A Practical Guide Addison-Wesley Educational Publishers, New York, 1997.

Return to Course Sylabi and Information Page