Comparing Survival Probabilities of Employers with Close Union Election Outcomes, Summaries of Business

Download Comparing Survival Probabilities of Employers with Close Union Election Outcomes and more Summaries Business in PDF only on Docsity! Do Unions Cause Business Failures?* John DiNardo University of Michigan, Ann Arbor and NBER David S. Lee UC Berkeley and NBER March 2003 Abstract Estimating the causal effect of unionization on business survival rates is difficult in the absence of large, representative data on establishments with union status information. It is also confounded by selection bias, because unions may tend to organize at highly profitable enterprises that are more likely to survive. Using new data on more than 27,000 establishments that faced organizing drives in the U.S. during 1983-1999, this paper utilizes a regression discontinuity design to estimate the impact of unionization on the probability of employer dislocation. Survival probabilities of employers where unions barely won the election (e.g. by one vote) are compared to those where the unions barely lost. The analysis yields a surprising result: little or no union effect on business dislocation rates over 1- to 18-year horizons. * An earlier version of the paper “The Impact of Unionization on Establishment Closure: A Regression Discontinuity Analysis of Representation Elections” is on-line as NBER Working Paper #8993, June 2002. We thank David Card, Larry Katz, Enrico Moretti, Morris Kleiner, participants of the University of Michigan Labor Workshop and the NBER Labor Studies Summer Institute for helpful comments and suggestions, Matthew Butler and Francisco Martorell for outstanding research assistance, Hank Farber for providing election data, and Christina Lee for reading previous drafts. 1 Introduction It is widely understood that unions raise the cost of labor by raising members’ wages above market rates.1 Unions also impose other costs on employers - limiting discretion in hiring and firing, for example, and altering the structure of pay differentials across skill groups. A key question for understanding the social costs of unionization is whether the wage premiums and other costs of unionism create large or small distortions in the allocation of labor.2 These distortions can take the form of reduced employment at unionized firms, or most dramatically, an accelerated pace of business failures. The potentially adverse effects of unions on firm survival are acknowledged by employers and employees alike. During union organizing drives, firms routinely threaten to close a plant if the union drive is successful [Bronfenbrenner 2000]. Employees seem to take these threats seriously: the risk of plant closure is cited as the leading cause of union withdrawal from organizing attempts [Commission for Labor Cooperation 1997]. Such risks are arguably higher now, in light of rapidly expanding trade with low-wage countries such as China and Mexico, and increasing international capital mobility.3 Despite the clear theoretical presumption and strong anecdotal evidence, the magnitude of the ef- fect of unions on establishment or firm survival is uncertain. One limiting factor is the absence of large, representative data sets that track establishments over time and provide information on union status.4 A second and even more important concern is the fact that unionization is nonrandom. Depending on the correlation between factors associated with higher or lower risks of survival, and higher or lower likelihood 1 See Lewis [1986]. 2 The presence of a “deadweight welfare loss” to unionization is a staple of textbook treatments of unionization. Even Freeman and Medoff [1984] - whose emphasis is on the possible “efficiency enhancing” aspect of trade unionism - stipulate the existence of such a welfare loss. More recently, the cross-country analysis of Nickell and Layard [1999] suggest that a change from 25 to over 70 percent of the workers covered by collective bargaining is associated with a doubling of the unemployment rate. Lalonde, Marschke, and Troske [1996], using a “difference-in-difference” approach with LRD data, find successful organization is associated with significant declines in subsequent employment and output. 3 The Commission for Labor Cooperation, a tri-national organization created under the North American Agreement on Labor Cooperation (“NAALC”) in response to labor issues related to NAFTA, called for a study on the impact of plant closings on union organizing in the three countries. 4 This has led researchers to use creative data collection methods to examine these questions. For example, Freeman and Kleiner [1990] conducted on-site interviews of 364 establishments that experienced representation elections in the Boston and Kansas City NLRB districts. Bronars and Deere [1993] construct a dataset of NLRB elections to COMPUSTAT data to construct a panel of 85 firms over a 20-year period. Freeman and Kleiner [1999] also use COMPUSTAT to construct a sample of 319 firms. Lalonde, Marschke, and Troske [1996] match NLRB representation elections to a subset of manufacturing establishments that are continuously operating in the LRD to create samples with 500 to about 1100 observations. 1 their economic implications in Section 6, and suggest directions for future research in Section 7. 2 Regression Discontinuity Analysis of Establishment Survival and Unionization: Basic Facts We begin by briefly describing our research design and summarizing our main results. Our regres- sion discontinuity analysis suggests that there is virtually no causal impact of unionization on survival rates of business establishments. The primary challenge of identifying the impact of unionization is one of isolating exogenous variation in the union “status” of an establishment, while keeping all other pre-determined characteristics of the establishment “constant”. A simple comparison of survival rates between unionized and non-unionized firms suffers from the potential endogeneity of union status. For example, firms with larger economic rents may be more likely to survive, and because of those rents, are more likely to generate worker demand for a union. In this paper, we exploit a distinctive feature of the union recognition process in the U.S. that we argue generates “as good as randomized” variation in union status. In the U.S. the obligation of employers to bargain “in good faith” with a union is almost alwasy determined through a majority secret-ballot vote amongst the workers; this “representation election” is overseen by the National Labor Relations Board (NLRB). We argue that it is plausible that there is at least some degree of unpredictability regarding the eventual vote tally, so that the firms and unions involved in elections where the union barely won (say, by one vote) are likely to be ex ante comparable to those firms and unions involved in elections where the union barely lost (by one vote). If this is true – that they are comparable “in all other ways” – then the difference in the survival rates subsequent to the election can be attributed to a causal impact of the union certification. Figure Ia graphically summarizes our main empirical finding. Using data on NLRB representation elections conducted between 1983 and 1999, it plots the proportion of employers that are still in business by the year 2001, by the actual share of the votes in favor of the union. Each dot represents an average 4 among elections within a 0.05 interval of the vote share. All points to the left of the 0.50 line represent establishments where the union lost the certification election, and those to the right represent union victories. If unionization substantially impacted ability of the employer to remain economically viable – a claim frequently touted by employers during organizing drives [Bronfenbrenner 1994, Kleiner 2001] – we would expect to see a sharp drop in survival rates at the 0.50 threshold. No drop is evident in Figure Ia. We interpret this smooth empirical relation through the 0.50 threshold as evidence that there is little or no causal effect of union status on employer survival. Two alternative explanations to the pattern in Figure Ia are 1) employers in which the union barely won and lost are not ex ante comparable, perhaps because there is no unpredictability in the outcome of the election, and that there is “sorting” of unions and employers on either side of the 0.50 threshold and 2) the intensity of the “treatment” is small among the employers and unions involved in close elections – that is, an employer that prevents union certification by 1 vote must make wage concessions as large as they would have had the union won by 1 vote. As we will discuss in greater detail in later sections, there are several reasons why the two interpre- tations are difficult to reconcile with the implications of a simple economic model of the election process, as well as with the empirical evidence we present. First, if the exact vote tally could be predicted with com- plete certainty, it is difficult to explain why a union would expend resources to ultimately lose an organizing drive. At the very least, if vote tallies could be perfectly predicted for a substantial fraction of the election cases (and if there were systematic “sorting” around the 0.50 threshold) we would expect to see a sharp drop in the relative frequency of observed elections in which the union barely lost an election. In fact, as Appendix Table II and Appendix Figure I show, union losses are at least as common as union victories, and the distribution of vote shares looks approximately normal, centered around 0.40, with no sharp drop in the relative frequency of bare union losses. More importantly, we present evidence below that employers and elections, in fact, look quite similar along many pre-determined characteristics on either side of the 0.50 threshold, as predicted by our implicit assumption of “as good as random” assignment. As one example, Figure Ic plots averages of our “union presence before the election” proxy – an indicator of whether a contract expiration notice 5 was filed prior to the election – against the election vote share for the union.8 The figure reveals no sharp change in pre-election “union presence” around the 0.50 percent threshold, suggesting that at least along this dimension, employers involved in close wins and losses appear comparable. Furthermore, the economic model we outline below illustrates that while the expected union wage could be expected to be higher given a higher probability that the union will win, we might also expect employers to offer a higher non-union wage in order to persuade some workers to vote against the union. Therefore, the gap between the expected union and non-union wage may not necessarily rise with the probability of a union victory. Actually, the minimum union-non-union wage gap needed in order to justify the unions’ costs of conducting an organizing drive must be larger for elections in which there is a lower probability of a union victory. Put simply, it would be unlikely that an election would be conducted if the union had nothing to gain from winning (i.e. if the expected “treatment” is zero). More importantly, we present evidence that is inconsistent with the hypothesis that union certifi- cation among close elections is ineffectual in altering union power. Figure Ib is analogous to Figure Ic, except we plot averages of our proxy for “union presence” subsequent to the NLRB representation election. It exhibits a striking discontinuity at the 0.50 threshold, suggesting that the certification has a significant causal influence in altering union power. Thus, it is evident in the data that something is “at stake” among these election cases. The remainder of the paper explains in greater detail the above reasons why we interpret our find- ings as evidence of little or no causal impact of unionization on employer survivorship. 3 Econometric Framework 3.1 Reduced-Form Regression Discontinuity Framework In the context of union representation elections, the internal validity of our regression discontinuity analysis primarily depends on two assumptions: 1) that the “treatment” (union recognition) is a known, discontinu- 8 The sample for the figure includes only those employers that have survived as of the year 2001. Below we discuss how to interpret these graphs in light of sample selection bias. 6 α is assumed to be negative, so that a higher wage leads to a higher probability that the employer is forced to shut down. Note however, that even if α is negative, given the necessarily non-linear relation between a probability andWj , it could very well be that the marginal effect ofWj on the probability of survival could be small. As in Ashenfelter and Johnson (1969), we assume that there are three separate agents, 1) the em- ployer, 2) the union leadership, and 3) the workers – the voters – in the potential bargaining unit. Manage- ment and the union each offer different levels of wages to the workers, and the workers vote for or against the union, based on those choices, so that we have Wj = WU j WINj +W N j (1−WINj) (5) WINj = 1 µ Vj > 1 2 ¶ where WU j and WN j are the wages offered by the union and management, respectively. This implies the relation y∗j = αW N j + α ¡ WU j −WN j ¢ WINj + εj (6) so that the causal effect of the union is represented by α ³ WU j −WN j ´ . Workers When considering whether or not to vote in favor of the union, the worker weighs the benefit of gaining a higher wage against the costs of a potentially higher probability that she will not retain employ- ment at the firm. A lower probability of retaining the job may arise either because the union will induce the employer to shut down or move, or induce the employer to scale back employment. Indeed, a worker may find the wage (and its consequences) offered by the management to be more reasonable and individually desirable. We describe the aggregate voting behavior of the workers as Vj = v ¡ WU j ,W N j ¢ + Uj (7) where v (·) translates the offered wages into a predictable component of the ultimate vote tally, and Uj is the unpredictable (by all agents in the model) component of the union vote share.12 12 Again, we could be more precise instead specifyingG−1 (Vj) = v ¡ WU j ,W N j ¢ +Uj , where G is a one-to-one transformation from the real line to the [0, 1] interval. 9 The shape of v (·) characterizes the workers’ preferences. We assume that ∂v ∂WU j and ∂v ∂WN j are both negative, capturing the notion that if the union raises the offer, it will lose the voters who are indifferent between the union’s and management’s offers. Similarly, the management can gain more “no” votes by promising a higher wageWN j . Employer The management seeks to maximize profits, weighing the benefits of offering a lower wage against the cost that lowering the wage induces a higher probability that the union will win, resulting in a higher wageWU j . The firm’s optimal choice ofWN j can be expressed as WN∗ j =argmax WN j H ¡ WN j ,W U j ,Pr [WINj = 1] ¢ (8) where H (·) is the employer’s objective function. An example of a specific form for the objective function is the expected profits, given the employer’s and union’s wage offers: H ¡ WN j ,W U j ,Pr [WINj = 1] ¢ = π ¡ WN j ¢ Pr [WINj = 0] + π ¡ WU j ¢ Pr [WINj = 1] (9) where π (·) is the profit function. Union The union leadership seeks to raise wages above that offered by the employer, but by offering higher wages, it reduces the probability of winning the election. The union’s optimal choice ofWU j facing the firm’s offerWN j can be written as WU∗ j =argmax WU j J ¡ WU j ,W N j ,Pr [WINj = 1] ¢ (10) where J (·) is the union’s objective function. An example of a possible form for this objective function is the expected net benefit (expressed in dollars) for the union: J ¡ WU j ,W N j ,Pr [WINj = 1] ¢ = −c+Pr [WINj = 1]U ¡ WU j −WN j ¢ (11) where c is a fixed cost to conducting an organizing drive, and U (·) is a benefit function with U (0) = 0.13 In a Nash equilibrium, in anticipation of how the workers will vote (on average), and given the correct expectation of the other party’s wage offer, union and management optimally choose their wage 13 In principle, U (·) could be decreasing in the wage gap at some point, if the union also gives weight to the employer’s survival and/or the level of employment. However, it is easy to imagine that in many election cases U (·) is increasing in the wage gap in equilbrium. 10 offers to maximize their objective functions. Implications It is clear from Equation 11 that if the outcomes of elections were known ex ante with certainty – if there were no Uj component – we should expect to observe no elections in which unions lose, if conducting an organizing drive is costly. In fact, as shown in Appendix Table II, over the sample period, the average vote share for the union and win rates are around 0.48 and 0.427, respectively. Furthermore, if in a large fraction of election cases, the outcome of a potential election were perfectly foreseen, we would expect to see a sharp drop in the relative frequency of a “close” union election loss. Appendix Figure I shows that empirically there is no such sharp drop; the distribution of vote shares is approximately normal, centered around 0.35 to 0.40. Our framework provides an intuitive explanation for these patterns: for every election, there is some ex ante probability (however small) that the union will prevail. In other words, there is some unpredictability in the final vote count. This would explain why unions would participate in elections that they so happen to eventually lose. Furthermore, presuming the existence of an unpredictable component Uj to the ultimate vote share, the framework above implies that the distribution of Uj would be continuous at Vj = v ³ WU∗ j ,WN∗ j ´ + Uj = 1 2 . This is because a discontinuity in the distribution ofUj at Vj = 1 2 is inconsistentwith the optimality of WU∗ j and WN∗ j for the union and management. This is because such a discontinuity would imply that Pr [WINj = 1] would be discontinuous in WU j and WN j at WU∗ j , WN∗ j .14 As evident from Equations 9 and 11, if this were the case, WU∗ j (WN∗ j ) would not be optimal, since the union (management) could lower (raise) wages by an arbitrarily small amount to affect a sharp rise (fall) in the probability of a union victory.15 Intuitively, no firm or union would settle with making a wage offer that could be altered a tiny amount in order to cause a discontinuous increase in the probability that the election would result in their favor. Thus, the existence of Nash Equilibria gives a theoretical justification of the validity of the regression discontinuity approach described above. 14 To see this, note that Pr [WINj = 1] = F ¡ v ¡ WU∗ j ,WN∗ j ¢− 1 2 ¢ , where F (·) is the cdf of −Uj . A discontinuity in the distribution of −Uj at Vj = 1 2 , implies that F (·) is discontinuous at v ¡ WU∗ j ,WN∗ j ¢ − 1 2 . As long as v ¡ WU∗ j ,WN∗ j ¢ is a continuous function of its arguments atWU∗ j ,WN∗ j , then Pr [WINj = 1] is discontinuous inWU j ,W N j atWU∗ j ,WN∗ j . 15 This is true as long as the profit function π (·) and benefit function U (·) are continuous. 11 propriate for an analysis of the impact of unionization on employer outcomes. First, that employers are thought to generally oppose organization drives [Kleiner 2001] suggests that both parties have “something at stake” in the outcome of the election. For example, we expect that both the union and management are expecting that a union win will generally lead to higher wages, more bene- fits, or better working conditions, at the cost of the employer. If very little were at stake, and if the elections themselves were pro forma events, then we would not expect to see a significant employer response to a union election victory. Such a finding would say more about the small size of the “treatment” (“union- ization”) than the potential magnitude of distortions caused by an aggressive union. This seems unlikely, however, since we analyze establishments which faced NLRB elections. Such a focus would seem likely to select establishments where union-management relations are contentious since in the overwhelming ma- jority of cases the management of such establishments always has the option of voluntarily recognizing the union without a (costly) NLRB election. Thus, it would seem more reasonable to assume that the outcome of the election is far from inconsequential to both parties. Second, combined with a contentious atmosphere, the secret-ballot nature of the vote undoubtedly generates a certain amount of uncertainty in the outcome of the election, particularly when the vote is expected to be close. As shown in Section 3 a certain degree of uncertainty is a crucial element to our theoretical and econometric framework. The assumption of some randomness to the vote would likely not be justified if union certification could be secured through a public petition. For example, if all that was required were 50 percent or more signatures, one could imagine that the sample of establishments/unions where the unions submitted a petition with 51 percent of the signatures would be very different from a (peculiar) group of establishments/unions where the workers submitted signatures that totalled 49 percent. By contrast, it is very easy to imagine in a secret-ballot context that those unions that won 26 out of 50 votes in a secret-ballot election possessed virtually the same ex ante chance of winning as unions that obtained 25 out of 50 votes (and lost). 14 4.2 Data Set Construction: the NLRB, FMCS, and InfoUSA, Inc. Deferring the details to the Appendix, we summarize here the most important features of the dataset used in the analysis. First, electronic records on all representation election cases handled by the NLRB in the fiscal years from 1984 to 1999 were obtained. These records have information such as the dates of the filing of the petition, the election, and the closing of the case, as well as the eventual vote tallies, as well as other characteristics such as the size of the voting unit, and the primary industry of the establishment in question. Importantly, these files contain the establishment name and exact address. The names and addresses alone were submitted to a commercial marketing database company called InfoUSA, Inc. InfoUSA main- tains an annually updated list of all active business establishments (with a telephone listing) in the United States. The basis for their database is the consolidation of virtually all telephone books in the country. InfoUSA makes a brief call to each establishment at least once a year, to verify their existence, and to update their information on various items such as 1) the total number of employees at the establishment, 2) the estimated sales volume of the establishment, 3) the primary product of the business, and various other characteristics. If InfoUSA found a record in their current database (as of May, 2001) with the same name and address, they appended their information to the record. They were not given any information beyond the name of the business and the street address. This merged data was then additionally merged to a database of all contract expiration notices between 1984 and February, 2001 – more than 500,000 case records – obtained from the Federal Mediation and Conciliation Service (FMCS) through a Freedom of Information Act (FOIA) request. According to the U.S. Code of Federal Regulations (29 CFR 1425.2) In order that the Service may provide assistance to the parties, the party initiating negotiations shall file a notice with the FMCS Notice Processing Unit ... at least 30 days prior to the expiration or modification date of an existing agreement, or 30 days prior to the reopener date of an existing agreement... Thus, in principle, parties to collective bargaining agreements are required to file so-called “30- day notices” with the FMCS. This was used to obtain our proxy for the “presence” of a union at the establishment – both before and after the election – under the presumption that contracts eventually expire, 15 typically after two or three years. There are a few important limitations to our data. First, our data do not constitute a true panel dataset. We only observe “survival” or “death” as of one point in time - in the year 2001. We know little about what happens between the time of the election and 2001, except the observation of contract expirations at that particular location. While we do observe a few “baseline” characteristics from the NLRB election file, InfoUSA does not retain historical records, so we do not have detailed employment and sales data for period between the election and the year we observe survival status (2001). Second, since we are measuring employer “survival” as a match (by name and address) in the InfoUSA database, there will undoubtedly be some measurement error. Consequently, we will inevitably treat some firms as having “died”, when instead we have simply been unable to match them. However, while this may mean that estimates of the level of survival rates may be downward biased, it is highly unlikely that establishments with close union winners are systematically less or more likely to match to the InfoUSA database than counterpart close union losers, except if there is a true impact of union certification on survival probabilities. Likewise, our measure of “union presence” will also likely be biased in levels, although this is unlikely to have important consequences for our comparison of close winners to close losers. For example, we understate the extent of unionization to the extent that our matching algorithm fails to locate a match in the FMCS data when such a match exists or to the extent that noncompliance with the law (regarding notifying the FMCS when a contract expires) is widespread. Alternatively, we overstate the extent of unionization to the extent that our matching algorithm produces “false positive” matches. Although the levels may be mis-measured, it seems reasonable to assume that these sources of measurement error are unlikely to be systematically different between close winners and close losers. On balance, we believe the benefits of being able to compare the bare winners and losers on the basis of some other measure of “union presence” other than the legal certification that necessarily results from winning the election outweighs the inability to obtain an accurate measure of the overall level of union presence. 16 Table II provides the details of these findings. The fifth row reports that employers where the union wins produce roughly 35 percent less in sales compared to cases where the union lost. Row (6) in Table II also shows that the outcome of the representation election is highly correlated with our proxy for union presence. This computation is made among the restricted sample of surviving (as of 2001) establishments. Among “union-loss” establishments we observe a contract expiring a approximately 10 percent of the time. When the union wins the election, on the other hand, there is a 36 percent chance that we will observe a union contract ending after the election. The rest of Table II provides good reason for the analyst to resist interpreting the union-won/loss differences in rows (1) - (5) as causal effects of union certification. For example, the establishments where the union won are about 15 to 20 percent smaller than the “union-loss” establishments, as measured by the number of eligible voters or the ultimate number of votes cast in the NLRB election. In light of these differences, it is thus not surprising that we observe differences in employment after the election, in the same direction, and of roughly the same magnitude. Similarly, row (7) of Table II reveals that we are more likely to detect the presence of a union before the election among establishments where the union won recognition compared to employers where the union lost: the respective proportions of pre-election “union presence” are 0.179 and 0.095. In addition, employers differ by election outcomes on a number of other dimensions; these differ- ences give more reason to maintain some doubt in any causal interpretation of the comparisons in rows (1) - (5) of Table II. For example, as row (12) of Table II indicates, establishments where the union won the election are much less likely (33 versus 42 percent) to be classified in the manufacturing sector. Moreover, as rows (13) and (14) of Table II indicate, establishments where the union won are more likely to be in the service sector (35 percent in “union-win” establishments versus 22 percent for “union-loss” establish- ments) and the voting unit less likely to be classified by the NLRB as “truck drivers”. On the other hand, measures of state economic conditions are not strongly related to the outcome of the election. The union won/loss differences in the levels and changes in the unemployment rate and the log(employment) level are statistically but not economically significant. 19 In sum, Table II provides evidence that caution the analyst against making inferences about the impacts of union certification on employer outcomes from simple differences in outcomes by election out- come.25 The evidence is suggestive that union recognition by election may be negatively selected – that unions are more likely to prevail in a representation election in smaller, and potentially less robust establish- ments. This would be consistent with the notion that larger establishments with greater resources may be more able to resist organizing drives. However, this interpretation of Table II is at best speculative without an independent estimate of the causal effect of union certification. 5 Estimates of the Impact of Unionization 5.1 Evidence on Validity of the Regression Discontinuity Design As mentioned above, if the regression discontinuity design is valid, employers involved in close union wins should be similar, on average, to those involved in close union losses, in terms of their pre-determined characteristics – whether or not they are observed by the econometrician. As in any empirical analysis, assessing whether “unobservables” are balanced is impossible. However, we can at least assess whether or not the regression discontinuity design is succeeding in balancing observable determinants of establishment survival. Our empirical analysis reveals that the examination of close elections does result in “treatment” and “control” establishments that appear to be otherwise similar on observable dimensions. Table III reports differences in characteristics of the employer, by union victory/loss, in the overall sample, and by sub-samples that isolate closer and closer elections. For example, in elections where the union lost, the average number of eligible voters is about 113, compared to about 92 where the union eventually won. That difference remains large when we examine elections where the union won between 25 and 75 percent of that vote. But it drops in half when we focus on the comparison among elections where the union won between 35 and 65 percent of the vote. The same holds true in percentage terms. The difference in terms of the log of the vote cast falls from -0.19 to about -0.10 when we move from the first 25 The researcher might be tempted to conduct the analysis conditional on the pre-determined characteristics such as industry, and size of voting unit, under the presumption that the election outcome is random conditional on those covariates. Besides being somehwat ad hoc, by “using up” the covariates, this approach has the drawback of eliminating any possibility of gauging the internal validity of the comparison. 20 to third set of columns. The differences in these average characteristics become even smaller when we focus our attention on all elections where the share of the vote for the union is between 0.45 and 0.50. For twenty-person votes, this means the outcome was decided by one vote. Table III shows that, for example, the differences in the pre-election size of the employer (as measured by the number of eligible voters and votes cast) fall to 2 or 3 (on bases of more than 100). Employers involved in union losses are more likely to be classified as manufacturing establishments in the overall sample (first set of columns), but that difference falls to a statistically insignificant -.025 when examining elections decided by the narrowest margin (the fourth set of columns). The monotonically decreasing differences, as one compares closer and closer elections, is also true for the proportion of employers categorized as service sector establishments, and for the proportion of voting units classified by the NLRB as “trucking”. Table III also shows that as one examines closer elections, the estimated standard errors rise, which would be expected, since the number of observations used for the analysis necessarily declines. At some point, restricting the sample to even closer elections will result in no observations for analysis. This il- lustrates the well-known trade-off between bias and variance in non-parametric estimation of an unknown conditional expectation function. Indeed, the averages in the last set of columns can be interpreted as kernel regression estimates using a uniform kernel of bandwidth 0.05. Insofar as the slope of the true conditional expectation function (of the variables with respect to the vote share) is nonzero, any kernel regression estimate will necessarily be biased in finite samples. A simple alternative way of using data points away from the 0.50 threshold to estimate population means at the threshold, is to specify a flexible-form parametrization of the underlying function. Figures IIa, b, and c illustrate the results from regressing the corresponding dependent variables on a fourth-order polynomial in the vote share, including a dummy variable Vj > 0.50, and the dummy variable interacted with a linear term in Vj . The predicted values of those regressions are super-imposed upon local averages by 0.05 intervals. Figure IIa, IIb, and IIc reveal 1) that the predicted values from the polynomial come reasonably 21 run could be canceling out a long-run negative effect (or vice versa), it is instructive to stratify the analysis by groups of years. The 2nd row of the first column presents the estimate from Column (6) of Table IV. The 3rd row reports that among elections that were held before 1988, the corresponding effect of union recognition on the probability of survival is about -0.022 with a standard error of 0.020. The following three rows report the interaction effects for the periods 1988-1991, 1992-1995, and after 1995, respectively. The interactions effects are small and statistically insignificant.26 Second, the effect could potentially vary significantly by industry. The next three rows in the first column of Table V show that the effects are slightly larger for service sector establishments, but again the interactions are not jointly statistically significant. Third, the effects could vary by the size of the voting unit (a rough proxy for initial size of the establishment). The final three rows of Table V show that the estimates are positive (0.005) for voting unit sizes between 20 and 40 workers, and are slightly negative (0.005-0.032=-0.027) for units between 40 and 100. But all interactions are not statistically meaningful at conventional levels of significance. The second column reports that the estimates of the overall effect and the various interactions do not change significantly, when we use an alternative measure of establishment survival. As mentioned in Section 4, we consider that an establishment has survived as of 2001 if the company name and address matches an entry in the InfoUSA database. However, in principle, if bare losers are much more likely than bare winners to undergo an ownership change – and hence change their name – then our primary measure of establishment survival may mask a true effect on establishment closure. A comparatively robust way to address this issue is to consider that an establishment has survived if any establishment is present at that exact address as of the year 2001 – irrespective of whether the company name changes. The second column of Table V reports the estimates from using this measure (Survival (2)), and shows that the estimates mirror that of the first column.27 On balance, while we cannot rule out small heterogeneous effects by these three observable di- 26 Specifically, the specification was the same as Column (6), except year dummies were replaced with time-period dummies, their interactions with theWINj indicator and their interactions with Vt. 27 Apparent from the first row, the obvious exception, as expected, is that the proportion of establishments that have any business (regardless of name) as of the year 2001, is significantly higher, at about 0.643. 24 mensions due to the magnitudes of the sampling errors, we interpret the estimates as indicating that our main estimate is not being wholly driven by a particular sub-sample, as defined along these observable dimensions. 5.3 Further Evidence on the Consequences of Union Recognition It is tempting to conjecture that a union that barely wins a representation election possesses the same degree of “bargaining power” as a union that barely lost. After all, even if the NLRA mandates that the employer bargain “in good faith” with the union as the exclusive representative of the workers in the unit, there is no guarantee that a first contract will be secured. Also, since nothing prevents a losing union from attempting another organizing drive (as long as one year has passed since the first election), one might conjecture that the outcome of a close electionmay in the long-run have no effect on the extent of union presence or power at the employer. As we discuss in greater detail in Section 6, this conjecture is difficult to reconcile with the simple economic framework we have outlined above. Among other reasons, it is difficult to explain why a union would undergo a costly organizing campaign if the outcome of the election was inconsequential for the union. More importantly, this conjecture is strongly inconsistent with our analysis of the data we have collected on post-election union presence. Specifically, while we do not directly observe the securing of a first contract, we observe the expiration of a first contract (and expirations of subsequent contracts). Thus, our post-election “union presence” variable is whether or not we observe at the employer’s location an FMCS contract expiration notice subsequent to the date of the election. The second to last row of Table III shows that there is a large difference in the probability of observing an expiration notice between union victories and losses. That difference becomes smaller, when examining closer elections, but remains a highly statistically significant 0.159 when examining elections with a union vote share between 0.45 and 0.55. In fact, our post-election “union presence” variable is the only variable in Table III where the differences remain large when examining the closest of elections. 25 The behavior of our post-election “union presence” proxy also stands out in our graphical analysis of the data. Figure Ib shows a smooth empirical relation between the vote share and the probability of observing a contract expiration notice – everywhere except at the 0.50 threshold, where there is a sharp jump in that probability from about 0.12 to 0.28. By contrast, the last row of Table III and Figure Ic, suggest no corresponding striking change for our pre-election “union presence” variable (an indicator of whether we observe a contract expiration notice at the address of the employer before the election). These figures suggest that the outcome of the representation election is far from inconsequential – that there is a permanent causal effect of a union victory on this particular proxy of union power. Also, the effect on our post-election union presence proxy is robust to alternative specifications, as shown in Table IV. This would be expected if the assumptions of the regression discontinuity design were valid. The estimates of the effect of union recognition on the probability that we observe a contract expiration notice at the establishment are precisely estimated and range from 0.132 to 0.152. Note from the comparison of Columns (3) and (5) or (4) and (6) that the inclusion of the pre-election union proxy does not meaningfully affect the union recognition effect, despite its own independent predictive power (t-statistic over 20).28 Some care needs to be taken in interpreting these findings in Table III and Figures Ib and Ic, because the averages are computed using the sample of establishments that have survived as of the year 2001. In principle, this induces a censored sample selection problem. In the same way that wages are not observed for the non-employed, our post-election union presence variable – and any other measure of union presence – is defined only for those surviving establishments29, there is therefore the potential that the discontinuity in Figure Ib is an artifact of sample selection bias [Heckman 1976]. However, in this particular context, the sample selection bias problem may not be important after all. Lee [2002] shows that if a treatment is “as good as randomized”, and if treatment affects sample selec- tion in a monotonic way, then equal probabilities of selection in the treatment and control group imply that there is no sample selection bias. These two conditions are applicable here: 1) the maintained hypothesis 28 Again, the coefficient on the pre-election union presence variable is not to be interpreted as a causal effect; rather it is more properly thought of as a partial correlation; its inclusion “absorbs” residual variation. 29 Actually, we do have some information for those that are not in business as of the year 2001, but we do not know when, between the date of the election and 2001, the establishment shut down. 26 prediction of our framework of optimizing employers and unions. Within our framework, it is possible to generate specific examples where such a prediction would hold, given specific assumptions about the shapes of the objective functions for the employer and union, the shape of v (·), and the distribution of−Uj . On the other hand, it is just as easy to generate the opposite prediction – that the gap in the wage offersWU j −WN j declines with an exogenous increase in the probability of a union victory. To see this, suppose that δj is a parameter that characterizes how much the workers care about wages so that we have v ³ WU j ,W N j , δj ´ where ∂v ∂δj > 0. Now suppose that the union leadership has a limited range of credible wage offers; to make the example stark, suppose that it is (correctly) expected by all agents that if the union wins, the wage WU j will be fixed at WU j . In this case, the equilibrium is determined solely by the management’s counter-offer wage. One can show that there are some reasonable assumptions under which ∂W N∗ j ∂δj > 0 and ∂ Pr[WINj=1] ∂δj , which would imply that the wage gapWU∗ j −WN∗ j diminishes with an exogenous increase in the probability that the union will win.31 Intuitively, faced with an exogenous increase in wage demands on the part of the workers (increase in δj), fearing the prospect of paying the high union wage, management offers a larger non-union wage in order to reduce the chance of losing the election to the union. In equilibrium, the probability of a union victory does increase, but it is moderated by the management’s optimal response. Therefore, in this case, we would expect to see a larger wage difference among close elections (where the probability of a union victory is more moderate) compared to elections in which the union overwhelmingly won (in which the probability was higher). There is another reason to expect that WU∗ j − WN∗ j may even be larger with a lower probability of the union winning: a union requires a larger minimum gain WU∗ j − WN∗ j in order to justify a costly 31 Let v ¡ WU j ,W N j , δj ¢ = −WU j − WN j + δj . Then the first-order conditions for the employer amounts to the equation f(v) 1−F (v) − ∂π(WN j ) ∂WN j π(WU j )−π(WN j ) = MB (v) −MC ¡WN j ¢ = 0, where f (·) and F (·) are the density and cdf for −Uj . The first term can be interpreted as the marginal benefit from raising the offered wage, which reduces the probability that the union will win, and the second term can be though of as the marginal costs (due to lower profits). An interior solution is guaranteed if f(v) 1−F (v) and ∂π(WN j ) ∂WN j π(WU j )−π(WN j ) are decreasing and increasing, respectively, in WN j . This will be true if −Uj is normally distributed and ∂2π(WN j ) (∂WN j ) 2 < Ã ∂π(WN j ) ∂WN j !2 π(WN j )−π(WU j ) . By the implicit function theorem ∂WN j ∂δj = ∂MB ∂v ∂MB ∂v + ∂MC ∂WN j which will be positive but less than 1. Therefore, an increase in δj leads to an increase in the offered wageWN j and in the equilibrium probability that the union will win. 29 organizing drive. This is immediately apparent from Equation 11, which shows that the expected benefit – itself a product of the probability and the benefit function U (·) – needs to exceed c to justify having an election in the first place. Put simply, the lower probability that the union will prevail, the larger wage gain is needed to justify the costs of organizing. For these reasons, it seems more plausible that the outcome of the election is far from inconse- quential to both the employer and the union, perhaps especially among close elections (where the ex ante probabilities of a union win are more moderate). However, whether or not the outcome of a representation election has any measurable impact on compensation and working conditions is ultimately an empirical question. Future research could show that the outcomes of elections are inconsequential in terms of alter- ing compensation and working conditions, which would suggest that the conventional wisdom regarding how unions affect wages would need to be re-evaluated. 6.2 Is the effect “economically” small? Our analysis obtains estimates around -0.01, and the null hypothesis of “no effect” cannot be rejected at conventional levels of statistical significance. However, it is also instructive to consider whether or not the estimates are consistent with an alternative null hypothesis – that the decline in the union sector in the U.S. in recent years is entirely attributable to the union impact on employer survival. Such “back-of-the-envelope” calculations require a great deal of abstraction. However, suppose we consider an economy initially made up of N0 establishments with identical, constant hazard rate of closure of d per year, and a constant inflow of bN0 establishments per year. Normalizing N0 = 1, it is straightforward to show that the number of establishments at time t is represented by n (t) = λ+ (1− λ) e−dt (12) where λ = b d , so that if b = d, the number of establishments remains constant over time. 32 Over the period from 1983 to 1998, union density among private sector workers fell by approx- imately 40 percent. The implied baseline hazard rate from our data of establishments implies a constant 32 This follows from a the differential equation n0 (t) = dn (t) + b. 30 hazard of d = 0.10.33 This implies λ = ¡ 0.6− e−0.10(15)¢ 1 1−e−0.10(15) = 0.485. Thus, if the decline of the union sector was entirely due to the union effect on the ability of establishments to survive (i.e. if the establishments were not unionized, λ would be 1), then the causal effect of union recognition would have to be a doubling of the hazard rate d. With a base hazard rate of 0.10, this implies that we should expect causal estimates of the 1-15 year survival rates to be on the order of -0.20.34 To the contrary, however, our causal estimates are around -0.01 and we are able to reject magnitudes larger than -0.04. Thus, our estimates cast considerable doubt on the proposition that the primary mechanism of re- cent union decline is through union effects on establishment closure. Of course, these rough calculations should be viewed with caution, as they are based on a number of compositional, aggregation, and behav- ioral assumptions – the empirical relevance of which can only be assessed with large-scale micro-data on establishments in the U.S. 6.3 The Impacts of Union Threat Our estimates of the impacts of unionization are conditional on having an NLRB election, and our model makes it clear that we are identifying the impacts of WU j −WN j on survival rates. As our model makes explicit, however, there is another potential impact of unions, which is the distortion due to “union threat” [Rosen 1969]. In other words, employers may face a higher closure rate because their own offered wage WN j is forced to be significantly higher than the competitive market wageWM j . There are two components to the “overall” impacts of unionization: one component is the effect that arises when wages are pushed fromWM j toWN j , and the other component – which we have estimated in this paper – results when wages are increased fromWN j toWU j . Since we find little evidence of effects on survival of the second kind, we conclude that evidence of large union effects on closure would more likely be found in analyses of union threat effects. However, our theoretical framework cautions against making the conjecture that threat effects 33 Actually, our data seem to fit a Weibull baseline better; however, as usual, it is difficult to know whether the empirical duration dependence is “real” or an artifact of unobserved heterogeneity. The 10 percent hazard rate is also roughly consistent with exit rate estimates from Dunne, Roberts, and Samuelson [1989] and McGuckin, Nguyen, and Reznek [1998]. 34 This can be seen by comparing the survivor function e−.1t and e−.2t over a 15-period interval. 31 in maximizing the profitability of the business – insofar as it would lead, for example, to a larger wage bill for the workers. Indeed, this possibility has been the focus of the “efficient contracts” literature (Mc- Donald and Solow (1981), Brown and Ashenfelter (1986), MacCurdy and Pencavel (1986), Card (1986), Abowd (1989), Abowd and Lemieux (1993)). The efficient contracts model departs conceptually from the monopoly union/“right–to–manage” model by observing that union members could value both the level of employment and the wage. And if the firm enjoys some economic rents – through monopoly power in the product market, for example – the monopoly union outcome is “inefficient” in the sense that at least one of the two parties can be made better off without making the other party worse off. In some situations – when there is a “strongly efficient” contract – there is a possibility that unions do not lower employment, but instead act to redistribute rents from firms to workers. 7 Conclusion This study meets two important challenges of credibly estimating the magnitude of the causal ef- fects of unions. First, we have constructed a large data set that represents a virtual universe of estab- lishments facing potential unionization, linked to comprehensive database on survivorship on businesses. Using over 27,000 observations, we have also exploited a feature of the NLRB election process to produce quasi-experimental estimates that are likely to be free of selection and omitted variable biases. Our results suggest that establishment closure is not the main mechanism of the employment reallocation response to unionization. Rather, it seems more likely that employers respond by reducing employment. In the existing literature, there is a fair amount of variation in estimates of “within-firm” employ- ment responses to unionization.38 It would seem fruitful to utilize this study’s research design to learn more about this margin of employment adjustment. There is also some uncertainty regarding the labor costs im- posed by unions. The existing literature suggests that union wage premiums are large, and that unions have 38 Estimates range from no effect to signficant effects. For example, Lalonde, Marschke and Troske [1996] suggest there is significant negative employment response. Card [1990] examines indexed and non-index union contracts and finds a significant negative employment response to higher real wages. On the other hand, Bronars and Deere [1993] examine a panel of publically- traded firms and finds no systematic relation between union growth and firm growth. Freeman and Kleiner [1990] find only modest employment effects (and modest wage effects) associated with new unionization. 34 significant effects on profitability [Ruback and Zimmerman 1984; Abowd 1989]. At the same time there is some evidence indicating only modest wage gains associated with new unionization [Freeman and Kleiner 1990; Lalonde, Marschke, and Troske 1996]. A useful avenue for future research is to use NLRB election data, as we have here, to explore the impact of unionization on: the level and distribution of wages, capital investment, and total factor productivity. 35 Appendix A. There are three sources of information that were merged for the analysis. We first describe how our data from the National Labor Relations Board (NLRB) was matched to firm data from InfoUSA. Next we de- scribe how the result of this matching process, henceforth the “NLRB/InfoUSA data” was matched to data from the Federal Mediation and Conciliation Service (FMCS). A.1 The NLRB to InfoUSA match First, electronic records on all representation election cases handled by the NLRB in the fiscal years 1984 to 1999 were obtained. These records contain information such as the dates of the filing of the petition, the election, and the closing of the case, as well as the eventual vote tallies, as well as other characteristics such as the size of the voting unit, and the primary industry of the establishment in question. Most importantly the file contains information on the name and street address at which the representation election was held. These 139,881 records were thenmatched by name and address to a commercial marketing database company called InfoUSA, Inc. Before being sent to InfoUSA, however, the address fields were first “stan- dardized" using a program called “Mailers+4 Postal Automation Software." For example, “1 Broad Street" was changed to “1 BROAD ST". This was done to facilitate matching the NLRB data to the data from In- foUSA. As discussed in the text, InfoUSA maintains an annually updated list of all business establishments (with a telephone listing) in the United States. The basis for their database is the consolidation of virtually all telephone books in the country. InfoUSA makes a brief call to each establishment at least once a year, to verify their existence, and to update their information on various items such as 1) the total number of employees at the establishment, 2) the estimated sales volume of the establishment, 3) the primary product of the business, and various other characteristics. We submitted the name and address information from our “address standardized" NLRB data to InfoUSA who matched as many of the submitted records to their current database (as of May, 2001) and then appended their information to the record. Apart from the name and address information, no other 36 Thus, we made a minor adjustment to the vote share variable in order eliminate this problem. For every case where there was an even number of votes cast, an amount equal to 0.5/(# votes cast) was subtracted from the vote share. For example, if 25 out of 50 votes were for the union, the vote share became 0.50 - 0.01 = .49; 26 out of 50 votes meant a vote share of 0.52 - .01 = 0.51. Cases where an odd number of votes cast were unadjusted. This minor adjustment restores symmetry in the support for vote share, and the new “vote share” variable still possesses the property that strictly more than 0.50 implies a union victory. Finally, the vote share was “binned” so that all vote shares between 0.50 and 0.55 were assigned the vote share of 0.525, shares between 0.45 and 0.50 were assigned the share of 0.475, and so forth. In this way, vote shares were standardized to the support for the elections with the smallest number of votes cast (20). A completely different approach is to abandon the use of the vote share completely, by focusing on the absolute vote count, and comparing elections in which the union either won or lost by literally 1 vote. This eliminates this “integer problem”, but at the same time tends to push larger establishments away from the threshold that determines victory (generating a pronounced U-shape in the average size of the establishment, with respect to the absolute vote margin of victory/loss). This was the approach used in DiNardo and Lee [2001]; it should be noted that the results reported there are qualitatively and quantitatively similar to the results in this paper, suggesting that our findings are not sensitive to the method used to address the integer problem. 39 References [1] Abowd, John M. “The effect of wage bargains on the stock market value of the firm” American Eco- nomic Review 79:774–800, September 1989. [2] Abowd, John M. and Thomas Lemieux. “The Effects of Product Market Competition on Collective Bargaining Agreements: The Case of Foreign Competition in Canada.” Quarterly Journal of Eco- nomics 108, (1993): 983-1014. [3] Angrist, Joshua D., and Victor Lavy. “Using Maimondies’ Rule to Estimate the Effect of Class Size on Scholastic Achievement.” Quarterly Journal of Economics 114 (1998): 533-75. [4] Ashenfelter, Orley and George E. Johnson. “Bargaining Theory, Trade Unions, and Industrial Strike Activity.” American Economic Review, 59 (1969) 35-49. [5] Boal, WilliamM., and John Pencavel, “The Effects of Labor Unions on Employment, Wages, and Days of Operation: Coal Mining in West Virginia,” Quarterly Journal of Economics 109 (1994) 267-298. [6] Bronars, Stephen G., and Donald R. Deere, “Union Organizing Activity, Firm Growth, and the Busi- ness Cycle,” American Economic Review, 83 (1993) 203-220. [7] Bronfenbrenner, Kate. “Employer behavior in certification elections and first contracts: Implications for labor law reform” In Sheldon Friedman, Richard Hurd, Rudy Oswald, and Ronald Seeber, editors, Restoring the Promise of American Labor Law, pages 75–89. ILR Press, Ithaca, New York, 1994. [8] Bronfenbrenner, Kate, Uneasy Terrain: The Impact of Capital Mobility on Workers, Wages and Union Organizing, research paper provided by the U.S. Trade Deficit Review Commision, 2000. [9] Brown, James N. and Orley Ashenfelter. “Testing the efficiency of employment contracts” Journal of Political Economy 94:S40–S87, 1986. [10] Campbell, D. T. “Reforms as Experiments.” American Psychologist 24 (1969): 409-29. [11] Card, David. “Efficient contracts with costly adjustment: Short run employment determination for airline mechanics” American Economic Review 76:1045–1071, December 1986. [12] Card, David, American Economic Review, 80 (Sep., 1990), pp. 669-688. [13] Commission for Labor Cooperation, Plant Closings and Labor Rights: A Report to the Council of Min- isters on The Effects of Sudden Plant Closings on Freedom of Association and the Right to Organize in Canada, Mexico, and the United States, 1997. [14] Davis, Steven J. and John Haltiwanger, “Gross Job Creation, Gross Job Destruction, and Employment Reallocation,” Quarterly Journal of Economics 107 (1992) 819-863. [15] DiNardo, John and David S. Lee. “The Impact of Unionization on Establishment Closure: A Regres- sion Discontinuity Analysis of Representation Elections,” Center for Labor Economics Working Paper #38, September, 2001. [16] Dunne, Timothy, Mark J. Roberts, and Larry Samuelson. “The Growth and Failure of U.S. Manufac- turing Plants.” Quarterly Journal of Economics 104 (1989): 671-698. [17] Freeman, Richard B. and James L. Medoff.What Do Unions Do? Basic Books, New York, 1984. [18] Freeman, Richard, and Morris Kleiner, “The Impact of New Unionization on Wages and Working Conditions,” Journal of Labor Economics, 8 (1990) S8-S25. [19] Freeman, Richard B. and Morris M. Kleiner “Do Unions Make Enterprises Insolvent?”, Industrial and Labor Relations Review, Vol. 52, No. 4, July 1999, pp. 507- 524. [20] Freeman, Richard B. “Why are unions faring poorly in NLRB representation elections?” In Thomas Kochan, editor, Challenges And Choices Facing American Labor. MIT Press, Cambridge, 1985. [21] Hahn, Jinyong, Petra Todd, and Wilbert van der Klaauw. “Identification and Estimation of Treatment Effects with a Regression-Discontinuity Design.” Econometrica 69 (2001): 201-209. [22] Heckman, James. “The Common Structure of Statistical Models of Truncation, Sample Selection, and 40 Limited Dependent Variables and A Simple Estimator for Such Models.” Annals of Economic and Social Measurement 5 (1976): 475-492. [23] Heckman, James J. “Sample selection bias as a specification error.” Econometrica 47 (1979): 153-62. [24] Kleiner, Morris. “Intensity of Management Resistance: Understanding the Decline of Unionization in the Private Sector” Journal of Labor Research, Volume XXII, Number 3, Summer 2001. [25] Kuhn, Peter, “Wages, Effort, and Incentive Compatibility in Life-Cycle Employment Contracts.” Jour- nal of Labor Economics, 4 (1986) 28-49. [26] Lalonde, Robert J., G. Marschke, and K. Troske, “Using Longitudinal Data on Establishments to Analyze the Effects of Union Organizing Campaigns in the United States,” Annales D’Economie et de Statistique 41/42 (1996) 155-185. [27] Lee, David S. “The Electoral Advantage to Incumbency and Voters’ Valuation of Politicians’ Experi- ence: A Regression Discontinuity Analysis of Close Elections” Center for Labor Economics Working Paper #31. UC Berkeley, April, 2001. [28] Lee, David S. “Trimming for Bounds on Treatment Effects with Missing Outcomes” Center for Labor Economics Working Paper #51, UC Berkeley, March, 2002. [29] Lewis, H. Gregg. Unionism and relative wages in the United States. University of Chicago Press, Chicago, 1963. [30] Lewis, H. Gregg. Union Relative Wage Effects: A Survey. University of Chicago Press, Chicago, 1986. [31] Macdonald, Ian and Robert M. Solow. “Wage bargaining and employment”. American Economic Re- view 71:886–908, 1981. [32] MaCurdy, Thomas and John Pencavel. “Testing the efficiency of employment contracts” Journal of Political Economy 94:S3–S39, 1986. [33] McGuckin, Robert H., Sang V. Nguyen, and Arnold P. Reznek. “On Measuring the Impact of Owner- ship Change on Labor: Evidence fromU.S. Food-Manufacturing Plant-Level Data.” in Labor Statistics Measurement Issues, J. Haltiwanger, M. Manser, and R. Topel, eds. Chicago, Illinois: University of Chicago Press, 1998. [34] Nickell, Steven, and Richard Layard, “Labor Market Institutions and Economic Performance,” Hand- book of Labor Economics, Vol. 3C, Orley Ashenfelter and David Card, ed. Elsevier Science, New York, 1999. [35] Pivetz, Timothy R., Searson, Michael, and James R. Spletzer, “Measuring job and establishment flows with BLS Longitudinal Microdata,” Monthly Labor Review, April 2001. [36] Rosen, Sherwin. “Trade Union Power, Threat Effects and the Extent of Organization.” Review of Eco- nomic Studies 36 (1969): 185-196. [37] Ruback, Richard and Martin Zimmerman, “Unionization and Profitability: Evidence from the Capital Market,” Journal of Political Economy, 92 (1984) 1134-57. [38] Thistlethwaite, D., and D. Campbell. “Regression -Discontinuity Analysis: An alternative to the ex post facto experiment.” Journal of Educational Psychology 51 (1960): 309-17. [39] Van der Klaauw, Wilbert. “Estimating the Effect of Financial Aid Offers on College Enrollment: A Regression-Discontinuity Approach.” Unpublished manuscript (1996). 41 3.00 3.20 3.40 3.60 3.80 4.00 4.20 4.40 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Local Average Polynomial Vote Share for Union Lo g( To ta l V ot es C as t) Figure IIa: Log of Total Votes Cast in NLRB Election, by Vote Share 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Local Average Polynomial Vote Share for Union Pr ob ab ili ty Figure IIb: Probability in Manufacturing (before Election), by Vote Share 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Local Average Polynomial Vote Share for Union Pr ob ab ili ty Figure IIc: Probability in Service Sector (before Election), by Vote Share 0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60 1.80 2.00 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 Density Normal Appendix Figure I: Estimated Density of Union Vote Share D en si ty Vote Share for Union Table III: Establishment Survival, Union Presence, and Pre-determined Characteristics, by Vote Share for the Union Full Sample 0.25<= Vote Share <= 0.75 0.35<= Vote Share <= 0.65 0.45<= Vote Share <= 0.55 Loss Won Diff. Loss Won Diff. Loss Won Diff. Loss Won Diff. Survival, 2001 0.430 0.400 -0.030 0.436 0.413 -0.023 0.438 0.424 -0.014 0.442 0.435 -0.007 (0.004) (0.005) (0.006) (0.005) (0.005) (0.007) (0.006) (0.007) (0.009) (0.011) (0.011) (0.015) Eligible Voters 113 92 -22 120 99 -21 118 106 -12 116 114 -2 (1) (1) (2) (1) (1) (2) (2) (2) (3) (3) (3) (5) Log(Elig. Voters) 4.29 4.14 -0.15 4.34 4.20 -0.14 4.33 4.25 -0.08 4.31 4.32 0.00 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.02) (0.02) (0.02) (0.03) Votes Cast 102 78 -24 108 87 -21 107 93 -13 105 102 -3 (1) (1) (1) (1) (1) (2) (2) (2) (2) (3) (3) (4) Log(Votes Cast) 4.18 3.99 -0.19 4.24 4.08 -0.16 4.23 4.13 -0.10 4.22 4.22 0.00 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.02) (0.02) (0.03) Manufacturing 0.421 0.326 -0.094 0.437 0.363 -0.074 0.436 0.372 -0.064 0.422 0.397 -0.025 (0.004) (0.004) (0.006) (0.005) (0.005) (0.007) (0.006) (0.006) (0.009) (0.010) (0.011) (0.015) Service Sector 0.218 0.348 0.130 0.223 0.322 0.099 0.234 0.315 0.081 0.251 0.290 0.038 (0.003) (0.004) (0.005) (0.004) (0.005) (0.006) (0.005) (0.006) (0.008) (0.009) (0.010) (0.014) Trucking 0.174 0.119 -0.055 0.162 0.128 -0.034 0.148 0.125 -0.023 0.135 0.124 -0.011 (0.003) (0.003) (0.004) (0.003) (0.004) (0.005) (0.004) (0.004) (0.006) (0.007) (0.008) (0.010) Number of Obs. 15818 11804 11851 8112 7351 5557 2217 1923 Union Present 0.097 0.363 0.266 0.105 0.344 0.239 0.117 0.332 0.216 0.117 0.276 0.159 Post-Election (0.004) (0.007) (0.007) (0.004) (0.008) (0.008) (0.006) (0.010) (0.011) (0.010) (0.015) (0.018) Union Present 0.095 0.179 0.084 0.097 0.155 0.058 0.101 0.149 0.048 0.099 0.132 0.033 Pre-Election (0.004) (0.006) (0.006) (0.004) (0.006) (0.007) (0.005) (0.007) (0.009) (0.010) (0.012) (0.015) Number of Obs. 6806 4726 5163 3350 3222 2358 980 836 Note: Standard errors in parentheses. Entries are means and differences by election outcome, in the full and, three separate sub-samples, by election vote share. Upper panel includes establishments, whether or not they survive as of the year 2001, and the lower panel includes only those establishments that survive as of the year 2001."Union Present post-election" (pre- election) indicates whether or not a union at the location of the establishment filed a contract expiration between the election date and 2001 (between the beginning of the FMCS data and the date of the election). Details of data set construction in the Data Appendix. Table IV: Reduced Form Specification: Effect of Union Victory on Probability of Survival (2001), and Post-Election Presence of Union (1) (2) (3) (4) (5) (6) (7) Probability of Survival Union Victory -0.007 -0.006 -0.003 -0.004 -0.008 -0.008 -0.007 (0.014) (0.014) (0.014) (0.013) (0.014) (0.013) (0.013) Presence of Union --- --- --- --- 0.094 0.093 --- (Pre-Election) --- --- --- --- (0.010) (0.010) --- Log(Eligible Vote) --- --- --- --- 0.022 0.012 --- --- --- --- --- (0.004) (0.004) --- Year Dummies No Yes Yes Yes Yes Yes --- State Dummies No No Yes Yes Yes Yes --- Industry Dummies No No No Yes No Yes --- Unit Dummies No No No Yes No Yes --- Probability of Post-Election Union Present Union Victory 0.152 0.150 0.143 0.145 0.131 0.135 0.132 (0.017) (0.017) (0.017) (0.017) (0.016) (0.016) (0.016) Presence of Union --- --- --- --- 0.401 0.394 --- (Pre-Election) --- --- --- --- (0.013) (0.013) --- Log(Eligible Vote) --- --- --- --- 0.015 0.006 --- --- --- --- --- (0.004) (0.004) --- Year Dummies No Yes Yes Yes Yes Yes --- State Dummies No No Yes Yes Yes Yes --- Industry Dummies No No No Yes No Yes --- Unit Dummies No No No Yes No Yes --- Note: N=27622 for upper panel, N=11532 (includes only those establishments that survive by 2001) for lower panel. Robust standard errors in parentheses. Upper panel refers to survival, as of 2001, as the dependent variable; lower panel refers to the observation of a contract expiration after the election. All specifications are least squares regressions, for the upper (lower) panel include a 4th order polynomial in the vote share, a dummy variable for win/loss and the dummy variable interacted with the linear term in the vote share. Column (7) regresses the residuals - from an initial regresssion of survival (presence of union) on all the covariates - on a 4th order polynomial in the vote share, a dummy variable for win/loss and the dummy variable interacted with the linear term in the vote share. Details of the data set in the Data Appendix. Table V: Reduced-Form Results: Impact of Certification on Establishment Outcomes, Overall, and by Year of Election, Industry, and Voting Unit Size Dependent Survival Survival (2) Empl. Log(Empl.) Sales Log(Sales) Variable Mean 0.417 0.643 83 4.42 14225 9.34 (Std. Dev) (0.493) (0.479) -276 -1.46 -51542 -1.67 Overall Effect -0.008 0.012 0 -0.07 250 -0.03 (0.013) (0.013) (8) (0.06) (1470) (0.07) Year: before 1988 -0.022 0.009 -17 -0.18 -326 -0.10 (Main) (0.020) (0.021) (10) (0.11) (1889) (0.12) Year: 1988-1991 0.001 0.003 13 0.17 -777 0.10 (Interaction) (0.028) (0.028) (14) (0.15) (2708) (0.16) Year: 1992-1995 0.034 0.013 26 -0.02 2065 -0.02 (Interaction) (0.029) (0.029) (16) (0.14) (3013) (0.16) Year: after 1995 0.022 0.003 34 0.25 1536 0.17 (Interaction) (0.029) (0.028) (17) (0.14) (3349) (0.15) Other Industry -0.009 0.018 -2 0.06 1641 0.06 (Main) (0.021) (0.020) (9) (0.09) (2087) (0.12) Manufacturing 0.002 -0.013 2 -0.24 -3887 -0.24 (Interaction) (0.025) (0.024) (12) (0.10) (2857) (0.13) Service -0.016 -0.004 3 -0.07 917 0.04 (Interaction) (0.027) (0.026) (17) (0.14) (2551) (0.16) El. Vote: < 40 0.005 0.023 -5 -0.11 2308 0.00 (0.022) (0.021) (10) (0.09) (1861) (0.12) El. Vote: 40-100 -0.032 -0.005 -4 0.02 -1223 -0.03 (0.026) (0.025) (11) (0.11) (2040) (0.13) El. Vote: > 100 -0.008 -0.028 30 0.06 -4318 -0.09 (0.028) (0.027) (18) (0.13) (3476) (0.16) Number of Obs. 27622 27622 26355 10265 25719 9629 Note: Standard Deviations in first row, robust standard errors otherwise. Survival(2) denotes whether any establishment was present at the exact street address as of 2001, third and fifth columns assign "0" to the "dead" establishments. Estimated Annual Sales Volume is in thousands of dollars. Specifications: Base specification is Column (6) of Table IV. See text for details on interaction specifications. See Data Appendix for details on data set construction