Although their clout has declined markedly over the past 30 years, unions remain a major influence in the lives of millions of Americans. One in eight Americans belongs to a union, and in 28 states workers can lose their jobs if they do not become a union member. Even in firms with voluntary membership, unions represent both unionized and non-unionized employees at the bargaining table, meaning it often makes sense for workers to join in order to have a say in the entity that negotiates their wages and workplace conditions.
When workers join a union, they get more than collectively negotiated terms of employment, however. They also get a leadership that pursues its own priorities outside of negotiating employment conditions, priorities which may or may not reflect union members' wishes. For all their influence in the workplace, it is not clear whether unions actually represent their members' values. While it is true that union members elect their leaders, union leaders appear to pursue an agenda disconnected from the concerns of their members.
One way to test if unions represent their members' priorities would be to give workers the option to choose whether or not to support a project backed by their union. If support remained at the same level, it would indicate that the union leadership's priorities matched those of the members. Decreased support would demonstrate that the unions' priorities do not reflect their members' wishes.
Just such an experiment has taken place over the last decade. Many unions are heavily involved in political activism, spending their members' mandatory dues to elect candidates favored by the union leadership. However, several states have passed "paycheck protection" laws that require unions to obtain written permission from their members before they can spend membership dues on political causes. In these states, the union's political activism becomes essentially voluntary for its members.
A detailed examination of union spending in states in which political donations are voluntary reveals that union leaders choose to spend far more on politics than their members would prefer. In particular:
Many union leaders are pursuing an agenda that their members do not support. This fact suggests that America's workers would be well served by giving them greater freedom in deciding whether or not to belong to a union, and union members should be given greater sway over how their dues are spent by their leaders.
Organized Labor Influential in Many Workers' Lives
Most of the news coverage of organized labor in recent years has focused on how dramatically union membership has fallen, or on discord in the labor movement and the splintering of the AFL-CIO. While unions are less influential than they were a generation ago, they still exert an enormous influence in the lives of millions of Americans. One in every eight American workers belongs to a union. In 28 states unions can negotiate contracts that require companies to fire workers who do not join the union. Even when a worker has the choice to opt out of union membership, it often makes sense to remain a member. Unions are legally the employee's sole representatives, representing a firm's union and non-union members alike. The only method a worker has of influencing critical decisions like strike votes, contract ratification, or negotiation strategies is to join the union and work to influence it from within. For many American workers, union membership is still either mandatory or their only practical choice.
Membership in a union involves more than voting to accept a contract or go on strike, however. Unions charge their members annual dues, averaging $125 per member in 2004. Unions use only a portion of that money to represent their members. America's 10 largest unions used an average of only 30 percent of their dues to represent their members and negotiate with employers. They spent the other 70 percent on activities like lobbying, political activities, gifts, grants, overhead, and compensation for their staff. These other activities might or might not reflect the priorities of the union's membership.
On the one hand, there is a strong presumption that unions pursue the priorities of their members because those members elect the union leadership. If they did not approve of the job their representatives were doing, they could simply elect new leaders. On the other hand, union leaders do little to inform their members of many of their activities, and it is doubtful whether union members would approve of many of the decisions of their leaders. Gerald McEntee, president of the American Federation of State and County Municipal Employees earns $580,000 a year-courtesy of the dues paid by his union's far less compensated members. The president of the National Education Association earns $439,000 a year. That is almost ten times what the average teacher earns. It is questionable whether union members really want their elected leaders to earn salaries several times larger than those paid to members of Congress. Unions also spent over $7 million in 2005 to hold conferences and workshops at resorts, including the Pheasant Run Resort & Spa near Chicago and the Palm Springs Riviera Resort in California, a use of funds which also may not reflect their members' priorities.
Misplaced union priorities extend beyond six-figure salaries and comfortable retreats for union bosses, however. Evidence suggests that the causes unions champion do not line up with their members' concerns. Unions have vocally campaigned against outsourcing and for stronger measures to improve job security. However only 9 percent of U.S. workers say that they fear their job will be shipped overseas, while workers say they are satisfied with their job security by an 82 percent to 15 percent margin. These numbers make little sense if unions are pursuing their members' agendas.
More evidence of a disconnect between union leaders' activities and their members' priorities comes from unions' political activities. Unions generously spend their members' mandatory dues on politics. In 2005, a non-election year, the AFSCME spent almost 20 percent of its budget on political activities and lobbying. The AFL-CIO spent 30 percent of its budget on the same causes. Despite the fact that almost two in five union members voted for President George W. Bush in the 2004 elections, both these unions gave over 97 percent of their donations to Democratic candidates. Do millions of union workers really want such a large portion of their dues to be spent on politics? Would Republican union members really want their dues spent to elect Democrats?
If so, then why would unions feel the need to mislead their members about where their dues go? The AFL-CIO Web site proclaims that the union's political program consists of encouraging "… union members to register to vote. We also research working families' concerns about current issues, and put together information showing where candidates for all levels of elected office stand on those issues." Nowhere does the site mention the hundreds of thousands of dollars in members' dues the AFL-CIO leadership has given directly to political candidates running in the 2006 elections. The AFL-CIO also implies that it does not spend mandatory union dues on political causes, claiming that "partisan political activities are paid for by voluntary donations from union members." This is not true. Union dues, not just voluntary donations, fund virtually every union's political agenda, including the AFL-CIO's. However, the AFL-CIO seems eager to hide this fact from its members, a decision which makes little sense if union leaders believe those members share their agenda.
Paycheck Protection Legislation Tests Union Leaders' Priorities
Anecdotal evidence suggests that union leaders pursue priorities that differ markedly from their members' concerns. If so, that would suggest that policymakers should either make it easier for workers to decline to join a union, or for union members to opt out of supporting those activities that they find objectionable. However, anecdotes are not facts. Although it appears unlikely, union members could well support the high salaries, policy concerns, and intense political activism of their leaders. To determine whether or not unions pursue priorities that their members support, there would have to be a situation in which union members had the option of declining to support their union on an issue while still remaining in the union. If support for that priority remained the same, it would demonstrate that-at least on that issue-the union accurately reflected the views of its membership. If support fell substantially it would show that-at least on that issue, and potentially on others-union leaders felt free to ignore the desires of their members when setting their union's priorities.
For the most part, union members do not have this option. Unions do not allow their members to selectively fund union conferences at the local Holiday Inn but not at the Soaring Eagle Casino and Resort in Michigan. Nor do union members have the option of declining to support a campaign to raise the minimum wage while backing their union when it fights for broader health care coverage. In almost all cases, members only have the choice to take it or leave it and cannot selectively back out of union activities they object to. However, a few states have passed legislation that gives workers the option of not funding their union's political activities. These laws provide an opportunity to test whether or not unions pursue their members' priorities.
Unions generously spend their members' dues on political activities, donating millions of dollars to like-minded candidates. In 2004 unions gave $61 million to federal candidates. Unions pay for virtually all of this political spending out of their members' mandatory dues. Under the Supreme Court precedent established in Communication Workers v. Beck (1988), workers cannot be forced to donate to political causes, and are entitled to demand a refund of the portion of their dues spent on politics. However, unions erect multiple roadblocks in front of workers who want to exercise their Beck rights. They implement bureaucratic obstacles, such as accepting such requests only 30 days out of the year, making it difficult for workers to formally request a refund of their dues. Often unions refuse to honor those requests unless workers file charges with the National Labor Relations Board. In many cases they also require members to resign from the union if they do exercise their Beck rights. So while workers theoretically have the ability to opt out their union's political campaigns, in practice that option is usually closed to them.
In response to these concerns, several states have passed what are known as "paycheck protection" laws to enforce workers' Beck rights. Rather than forcing workers to navigate union roadblocks to obtain their dues refund, paycheck protection laws require unions to obtain annual written consent from their members to spend their dues on political causes. The union must refund the portion of dues spent on politics of any member who does not agree in writing to have his or her dues spent for that purpose. Thus paycheck protection allows every member of the union who wishes to engage in political activism to do so, while ensuring that workers who either object to that activism or would rather spend their money in other ways have the freedom to opt out.
The existence of these laws provides the opportunity to test how closely union leaders hew to their members' priorities. If analysis shows that union members continue to support their leaders' political campaigns when doing so becomes completely voluntary, it would show that union leaders are simply reflecting the desires of their members. On the other hand, a significant decrease in union campaign contributions would demonstrate that unions pursue an agenda and priorities that differ from those of their members. To test union responsiveness to member concerns, this paper undertakes regression analysis to determine what happens to union campaign donations after workers have the choice of opting out.
Data Sources and Description
Six states have passed paycheck protection legislation, although in Ohio, courts enjoined the law and it never took effect. Washington State, Michigan, and Wyoming passed paycheck protection laws that applied to all unionized workers in those states, while Idaho and Utah passed paycheck protection laws that applied only to public sector unions. This report focuses on the effects of paycheck protection on public sector union campaign donations to state candidates in order to maximize the number of states included in the analysis.
Campaign finance data come from the Institute on Money in State Politics, a non-partisan, non-profit organization that compiles a database of state campaign donations. The Institute provides data from state campaign-contribution disclosure forms, and records the source and size of donations to candidates for public office. (The Institute does not have data on campaign spending by outside interest groups, only on donations made directly to campaigns.) The Institute's data allow the construction of a panel dataset of campaign contributions made by public sector unions to candidates for state office by state and election cycle.
The Institute on Money in State Politics initially began as a project to study campaign finances in select western states; it received additional funding to expand into a national organization in the late 1990s. Consequently the geographic scope of the data available from the Institute increases over time. The Institute has information on campaign donations by public sector unions for only nine states in the 1990 election cycle. Its coverage expands to all states by the 2004 election cycle. Additionally, the number of races covered in each state expands over time. In Utah, for example, the Institute had data on donations to candidates for the state legislature only in the 1990 election cycle. By the 2004 election, it also had information on donations to candidates for governor, other statewide offices, and judicial posts.
This introduces a spurious upward trend in total campaign donations caused by the Institute's ability to examine more races, not by a true increase in campaign spending. This also makes campaign finance data not directly comparable between states, as the reports record the total amount given to campaigns for different races. To correct this problem, Heritage Foundation researchers collected campaign contribution data for only state legislative races, even when information about races for other offices existed. In the case of Utah, for example, only information on contributions to candidates for the state House and state Senate were collected for each election cycle, although by 1998 data on other state races were available. In this way the magnitude of total campaign contributions are directly comparable across states and election cycles.
This results in an unbalanced panel dataset of state-level campaign donations made by public sector unions to state legislative candidates, with individual observations at the state-election cycle level. Data on donations made in 18 other sectors of the economy were also collected for use in robustness checks on the results, which will be presented later. In order to make meaningful comparisons between campaign donations in large and small states, these data are expressed in terms of total spending per 1,000 residents using annual population data from the Bureau of Economic Analysis (BEA). All monetary figures are also adjusted for inflation.
To account for paycheck protection laws a variable is created that is set to 1 if a paycheck protection law is in effect in that state during that election cycle and set to 0 otherwise.
Factors other than paycheck protection laws can be expected to affect campaign donations and should be accounted for in any analysis. States with greater proportions of union members probably see more union spending, so the proportion of the workforce of each state that belongs to a union is included as a control. Similarly, states with larger public sector workforces probably see more donations by public sector unions, so BEA data on the proportion of personal income in each state spent on state government wages and salaries are included in the analysis.
States with wealthier residents may see more campaign contributions because potential donors have more to give, so personal income per capita data from the BEA are included for each state-election year. Election spending may rise during gubernatorial or presidential election cycles. Even though only state legislative races are examined, a dummy variable set to 1 if a presidential or gubernatorial election occurred that cycle and 0 otherwise was added. Similarly, variables were added indicating the proportion of each state legislative chamber up for re-election that cycle. Demographics may account for campaign contribution levels, so the proportion of state residents who are white and the proportion with bachelor's degrees or higher were included. Finally a year variable was included to account for potential time trends in campaign contributions over election cycles.
Appendix A contains a table of descriptive statistics for these variables.
The basic specification of the model is presented in equation 1. For each state s and cycle t we have
The union contribution data are expressed in logs to ease the interpretation of the results. When the dependent variable is a log value, then the regression coefficients on the independent variables reveal the percent change in the dependent variable associated with a unit change of the independent variable. So β1 reveals the percent change in public sector union contributions associated with the enactment of paycheck protection laws.
Two different models are used to estimate β1: the fixed effects model that eliminates fixed differences between states and the random effects model that does not. For completeness both results are reported, although econometric tests show that the fixed effects estimator is preferred. Appendix B explains the differences between these estimators in greater detail.
Table 1 reports the results from the basic specification in column A. Paycheck protection laws have a clear and negative effect on public sector union campaign contributions to state legislative candidates. This effect is highly statistically significant. Paycheck protection laws are associated with reductions in public sector union donations of over 50 percent. The odds of this reduction resulting from random chance and not a true correlation with paycheck protection laws are less than 1 percent. It appears that union members opt out en masse when they have the option of not funding their leader's political priorities.
However, all but one of the states that passed paycheck protection laws are in the western United States. It could be that unions in western states are weaker and give less to candidates for state office than do those in other states. Perhaps what looks like the effect of paycheck protection laws actually reflects regional differences in campaign contribution levels. To test this possibility the regressions were run again with only western states included in the analysis. The results from these regressions are reported in column B of Table 1. The results are broadly similar to those reported in column A. The magnitude of the reduction in public union donations rises slightly, to a decrease of over 60 percent. It appears that the correlation between lower union campaign contributions and paycheck protection laws does not simply reflect a regional effect.
Correlation does not necessarily imply causation, however. Large numbers of sick people reside in hospitals, but that does not mean hospitals make people sick. Similarly it could be that instead of paycheck protection laws causing union spending to drop, paycheck protection laws are passed in states where organized labor has little influence and spends less anyway.
A number of states have come close to passing paycheck protection laws, or in the case of Ohio, passed them and seen them struck down by the courts. If paycheck protection laws pass when unions are already weak, that would suggest unions are also weak when paycheck protection laws come close to passing. Column C presents regression results where attention is restricted to the five states that passed paycheck protection and the four states that came close to enacting paycheck protection laws. If paycheck protection is simply a result and not the cause of decreased union expenditures, then the effect ought to disappear when the only states studied are those where unions are weak. As Table 1 indicates, paycheck protection laws still show statistically significant decreases in public sector union spending on the order of 50 percent.
It could be the case that the states which passed paycheck protection saw fewer political donations overall after the law was passed for reasons that had nothing to do with the legislation itself. These estimates cannot rule this out. To test this possibility Table 2 presents regression results examining the effects of paycheck protection laws on campaign contributions from the 18 economic sector divisions used by the Institute for Money on State Politics.
There are obvious reasons why paycheck protection laws would affect union campaign contributions. But it is hard to see how they would affect contributions from most other sectors of the economy, such as donations from farmers or businesses. If paycheck protection laws are associated with changes in campaign contributions in areas where they would be expected to have no effect, it would suggest that something other than the new laws might be causing public sector union donations to fall.
Table 2 shows that this is not the case. The vast majority of the estimates are statistically insignificant. Most of those that are significant are in sectors where a causal effect of paycheck protection laws makes sense.
The results show that paycheck protection laws have a significantly negative impact on contributions from labor unions. This does not come as a surprise: 48 percent of all union members in the United States work for the government, and three of the five enacted paycheck protection laws also applied to private sector unions. This finding simply confirms the previous results. They also have a large negative effect on contributions from unknown sources, though these unknown sources could well be poorly reported contributions from unions.
Three other estimates in Table 2 are statistically significant:
the random effects estimate of candidate contributions to
their own campaigns and both the random effects and fixed
effects estimates of contributions from lawyers and lobbyists.
These are sectors which should not be affected by paycheck
protection laws. However, Table 2 presents 36 coefficient
estimates. Random chance will guarantee that three to four of those
estimates will show statistical significance at the 10 percent
level, even when no true correlation exists. Therefore the
statistical significance of donations from these sectors- sectors
where a causal connection with paycheck protection laws is hard to
see-probably results from normal statistical variation. The
overwhelming majority of the estimates show no statistically
significant effect. It appears that paycheck protection laws do not
reduce contributions from doctors, farmers, real estate agents, or
most other donors. They do reduce donations from organized labor,
the one sector of the economy directly affected by the laws.
These regression results provide strong statistical evidence that paycheck protection legislation reduces campaign contributions by public sector unions. Appendix C discusses additional robustness checks performed to ensure that these results are reliable. Given the choice, it appears that workers would rather spend their money on something other than their union leadership's political ambitions.
Effect on Union Political Influence
Although union campaign contributions fall by approximately 50 percent when workers get a say in how their money is spent, that does not necessarily mean that union political activity also drops. The Institute on Money in State Politics collects data only on how much money unions give to other candidates or to political parties, so-called hard money. It does not provide any information about how much unions spend directly on politics. Many of the state paycheck protection laws permit unions to spend their own money on advertising campaigns near elections, get-out-the-vote drives, and otherwise campaign directly, so long as they do it themselves.
No database of union spending of so-called soft money exists, so it is impossible to say if paycheck protection laws actually prohibit union leaders from spending dues on priorities their members do not share. Circumstantial reports certainly suggest that union leaders simply ramp up their unconstrained soft money spending when their members have the option of opting out of hard money donations. The Washington State teachers union responded to the paycheck protection law passed in 1992 by sending members' mandatory dues to the Community Outreach Program, from which teachers did not have the choice of opting out. The COP then spent millions of dollars of teachers' dues on political causes, effectively sidestepping the law.
Unions may well have found ways to circumvent the intent of most of the paycheck protection laws passed by the states, finding ways to spend their members' dues on politics even when their members object. There is simply not sufficient information on union soft money spending to draw conclusions about whether or not paycheck protection reduces union political activity. What is clear is that paycheck protection laws reduce the direct union contributions to candidates.
Many American workers belong to unions as a matter of necessity. Some would lose their job if they left their union; others join because they want to participate in the union that negotiates their wages and benefits. All of these workers pay mandatory union dues that are spent by the union leadership. The priorities of these union leaders often seem out of step with those of their members. Six-figure salaries, activism on causes that do not concern most workers, and heavy political involvement do not appear to resonate with most union members.
Paycheck protection laws, which require workers to agree to have their union dues donated to political campaigns, provide an opportunity to test how closely unions reflect their members' wishes. The results are clear and unambiguous. Union campaign spending falls by 40 to 50 percent once workers get a say in how their dues are spent. Unions are not spending their members' mandatory dues in accordance with those members' wishes. Unions may well use loopholes to get around these restrictions, but the fact that they need to use loopholes shows how out of touch they are with their members' priorities.
Union leaders are pursuing an agenda that union members do not support. Union members need to have more options in order to ensure that unions pursue their priorities. Congress and the states should pass more effective paycheck protection laws that close the loopholes unions have used to continue spending their members' dues even over their objections. Workers who find that unions still do not pursue their priorities should not be forced to belong to a union in order to keep their job. Workers should not be forced to serve their union's priorities, but unions the priorities of their workers.
James Sherk is a Policy Analyst in Macroeconomics in the Center for Data Analysis at The Heritage Foundation. The author thanks Heritage Foundation interns Ben Keefer and Kevin Kellert for their invaluable contributions to this paper.
Fixed Effects vs. Random Effects Estimators
When conducting regression analysis it is important to be aware that the error terms may violate traditional assumptions. The error terms will not be independently identically distributed if there are other unobserved state-specific factors that affect campaign spending but were not included in the control variables. In that case, the error terms correlate across the same state over time. If, for example, public sector unions have less social support in a heavily conservative state like Idaho, then their donations will probably be smaller than average. These factors could interfere with the regression analysis.
Using panel data provides an opportunity to control for state-specific factors such as this, as long as those factors are fixed over time. Econometricians use two basic models to correct for this potential problem. The first is a random effects model (RE). Under the assumption that the state-specific effects are uncorrelated with the explanatory variables, the random effects estimator provides an efficient estimate of the coefficient. In other words, it provides the estimate most likely to distinguish a true correlation in the data from statistical noise.
However, the assumption that the unobserved effect has no correlation with the other explanatory variables is rather heroic. Consider the case of Idaho again. It may well be that more conservative states with smaller public sectors (and thus public sector unions) are more likely to pass paycheck protection laws. In that case the unobserved effect of being heavily conservative will be correlated with the passage of paycheck protection laws, but also with lower union contributions, which could give the appearance that paycheck protection laws lower union spending when there is no true effect.
In the case where the unobserved effect is believed to be correlated with the explanatory variables, the fixed effect model (FE) is appropriate. This estimator subtracts from each state observation its average value over time. Removing the average removes any fixed effect in the data and the FE estimator is unbiased even when state-specific fixed effects are correlated with the explanatory variables. However, the FE estimator is less efficient than the RE estimator, making it harder to claim a correlation is statistically significant.
It should be noted that the FE and RE are identified off of different sources of variation. Most of the variance that the RE estimator exploits comes from the difference in contribution levels between states that passed paycheck protection laws and those that did not. By contrast most of the variation the FE estimator utilizes comes from the difference in contribution levels over time in the states that passed paycheck protection laws.
To test whether the RE or FE model is an appropriate specification, econometricians use a Hausman test. This test compares the coefficients from an efficient but potentially biased model (the RE estimator) and compares them with the coefficients from an unbiased but inefficient model (the FE estimator). The test indicates whether or not the coefficients from the two models differ systematically from each other. If they do not, that supports using the more efficient RE estimator. If they differ, that indicates that bias does affect the RE model and an unbiased but less efficient estimator should be used.
A Hausman test was performed on the paycheck protection coefficient in the RE and FE models presented in Table 1. The test rejects the null hypothesis of no systematic difference between the coefficients, with a p-value of 0.057. This supports the FE specification. For the sake of completeness, however, RE results are also presented throughout this report.
Additional Robustness Checks
In addition to the results presented in the main body of the paper, additional robustness checks were performed to ensure the reliability of these results.
Dropping One State at a Time
Since only a few states have passed paycheck protection laws, these results could be driven by an outlier while paycheck protection had little effect in the other states. To ensure this did not account for the results, the regressions were re-run, omitting in turn each state that had passed paycheck protection laws from the sample. If only one state explained the results, this procedure would reveal that.
The results from these regressions were broadly similar to those performed previously and were statistically significant at the 5 percent level in every case. They are presented in Columns 1 through 5 of Table 4. The point estimates were slightly more dispersed than previous estimates, but still clustered around a 40 to 60 percent decrease in union donations after the paycheck protection laws passed.
The panel dataset used is unbalanced, with more states being added over time as the Institute on Money in State Politics received funding to examine additional states. There should be no relation between the Institute's fundraising and states passing paycheck protection laws, so the unbalanced nature of this panel should not affect the final results. Nonetheless, to test to ensure that the results held with a balanced panel the basic FE and RE regressions were run on the reduced set of states for which data on public sector union contributions were available in every election cycle between 1990 and 2004.
Column 6 of Table 4 presents the results from this regression. The RE results are significant at the 1 percent level. The FE results were barely significant at the 10 percent level. However, the reduced significance is hardly surprising when the sample size has been reduced by three-quarters. These results are taken as evidence that using an unbalanced panel does not drive the estimates presented here, but that these results still hold when using a balanced panel.
The robust standard errors generated by the STATA software used to calculated these results do not correct for serial correlation. I re-estimate the model using a heteroskedasticity and auto-correlation robust OLS estimator, the results of which are presented in Column 7 of Table 4. The results are significant at the 1 percent level, and similar in magnitude.
Two other robustness tests were performed. Time effects were modeled with a linear year variable. Column 8 of Appendix C shows the results when eight election cycle dummy variables were used in place of a linear time trend. The fixed effects estimator remains statistically significant at the 3 percent level, while the random effects estimate is now statistically insignificant. As the Hausman test rejects the RE estimator in favor of the FE estimator, this is not taken as a cause for concern.
As a final test, each explanatory variable besides paycheck protection was omitted from the sample, and the regressions were run again. In all cases the paycheck protection variable remained significant at conventional levels.
 Based on dues paid by the 10 largest unions by membership in the United States. The data come from LM-2 forms filed with the Department of Labor and are available from the author upon request. The unions included in this list are theUnited Food and Commercial Workers, the Service Employees International Union, the International Brotherhood of Teamsters, the United Auto Workers, the Laborers International Union of North America, the National Education Association, the American Federation of Teachers, the American Federation of State and County Municipal Employees, the International Brotherhood of Electrical Workers, and the Communication Workers of America.
 Author's calculations based on data compiled from the Department of Labor by UnionFacts.com at http://www.unionfacts.com/ unions (July 5, 2006). Note that the Communication Workers of America did not report the proportion of dues spent on representational activities and so this figure is the average value for the remaining nine unions.
 Data from UnionFacts.com, based on data provided by the Department of Labor, at http://www.unionfacts.com/unions/unionProfile.cfm?id=289 (July 5, 2006).
 Data from UnionFacts.com, based on data provided by the Department of Labor, at http://www.unionfacts.com/unions/unionProfile.cfm?id=342 (July 5, 2006).
 Teacher pay averaged $47,808 in 2005. "Inflation Outpaces Teacher Salary Growth in More Than 40 States," National Education Association Press Release, December 5, 2005, at http://www.nea.org/newsreleases/2005/nr051205.html (July 5, 2006).
 Karlyn H. Bowmanand Bryan O'Keefe,American Enterprise Institute Public Opinion Study, August 30, 2005, at http://www.aei.org/publications/pubID.19072,filter.all/pub_detail.asp (July 5 2006).
 Data from UnionFacts.com, based on data provided by the Department of Labor, at http://www.unionfacts.com/unions/unionProfile.cfm?id=289 and http://www.unionfacts.com/unions/unionProfile.cfm?id=106 (July 5, 2006).
 Ibid. Exit poll results as reported by CNN at http://www.cnn.com/ELECTION/2004/pages/results/states/US/P/00/epolls.0.html (July 5, 2006).
 See PoliticalMoneyLine at http://www.tray.com/cgi-win/x_pacpg.exe?DoFn=C0000380606 (July 5, 2006).
 AFL-CIO, About Us: Union Facts.
 Data from OpenSecrets.org, based on FEC filing data. See http://www.opensecrets.org/industries/indus.asp?Ind=P (July 6, 2006).
 Robert Hunter, "Paycheck Protection in Michigan," The Mackinac Center for Public Policy, September 1998, pp. 6-7, at /static/reportimages/FC951BF881B8886CC19CD5C6A6DBBA6C.pdf (July 6, 2006).
 Ibid., p. 5.
 Note that five states did not hold elections to state office in the 2004 election cycle: Alabama, Louisiana, Mississippi, New Jersey, and Virginia. Alabama holds elections to the state legislature every four years. The other four states hold their elections in odd-numbered years. Since no state that passed paycheck protection legislation holds elections to state office in an odd-numbered year, the data were restricted to include only even-numbered election cycles.
 State legislative races were selected for two reasons. First, they were typically among the first races for which the Institute collected data, so using them increased the number of useable data points. Second, most states hold legislative races every election cycle, making them a more consistent measure of campaign donations than contests for gubernatorial or other statewide offices, which are typically held every other election cycle.
 The Consumer Price Index for All Urban Consumers-Research Series was used to express all monetary values in 2004 dollars.
 The paycheck protection variable is set to 1 for the following state-election cycles. Idaho: 1998, 2000, 2002, 2004. Michigan: 1998, 2000, 2002, 2004. Utah: 2002, 2004. Washington State: 1994, 1996. Wyoming: 1998, 2000, 2002, 2004. Washington State's law remains on the books, but is essentially ineffective after an agreement between the state's largest teachers union and the Washington State Attorney General in early 1998, and is thus coded as a 0 for cycles after 1996.
 Data on racial composition and educational attainment came from the Census Bureau. The author was unable to locate education data for 1992 for use in the 1991-1992 election cycle, and so used data from 1991 instead.
 To see this, note that when log(Y) = Xβ + ε we have that the derivative of the log of Y with respect to X is given by β = d[log(Y)] / dX = dY / Y / dX and that dY / Y / dX is the percent change in Y given a unit change in X.
 The log specification is presented here solely for ease of exposition and does not materially affect the results. Calculations performed using the level, not the log, of campaign contributions are available from the author upon request.
 Note that observations are weighted by the square root of the state population size. Alternative specifications tested included using the population size as weights, and not weighting the data. Neither alternative specification materially affects the results presented here.
 The states included in this analysis were Arizona, Colorado, Idaho, Nevada, New Mexico, Oregon, Utah, Washington, and Wyoming.
 Voters in California and Oregon defeated paycheck protection ballot initiatives in 1998, with 53 and 51 percent "No" votes, respectively. Nevada had a paycheck protection initiative on the state's ballot which was leading in the polls before being removed from the ballot by the courts. Ohio enacted a paycheck protection law in 1995, and again in 2005, but court rulings prevented it from taking effect.
 Results were omitted for public subsidies for campaigns. No state recording taxpayer contributions to campaigns has also enacted paycheck protection legislation, and thus identification of a paycheck protection effect on public subsidies is impossible.
 Andy Furillo, "Unions Have Blunted Bids to Curb Political Spending," The Sacramento Bee, June 27, 2005, p. A1, at http://www.sacbee.com/content/politics/story/13137532p-13981621c.html (July 11, 2006).
 Michael Reitz, "Paychecks Unprotected: Lessons Learned From California and Other States," Capital Research Center Labor Watch, January 2006, at /static/reportimages/D9519CD091940FE868E690C5897C1835.pdf (July 11, 2006).
 For additional information on random effects and fixed effects estimators see Jeffrey Wooldridge, Econometric Analysis of Panel and Cross Section Data (Cambridge, Mass.: The MIT Press, 2001), Chapter 10.
 These states are Alaska, Idaho, Montana, Nevada, Oregon, Washington State, and Wyoming.
 Note that while the use of logs or levels of campaign contributions does not generally affect the results presented in this paper, that is not the case here. The FE estimates of the level of contributions in the balanced panel sample are only significant at the 18 percent level, below conventional significance thresholds.
 The RE estimator is only statistically insignificant when the log of public sector union contributions is used. Both the FE and RE estimators are significant at the one percent level when the level of public sector union contributions is used.