Erin O’Hara O’Connor
Florida State University College of Law
Kenneth J. Martin
Randall S. Thomas
III. Univariate Analysis
We begin our empirical analysis with a description of our data- collection procedure and an overview of the prevalence of arbitration clauses in our sample. We then present some univariate statistics concerning the use of arbitration clauses, followed by a discussion of the more refined forms of customization of these clauses. We conclude this section with an analysis of the use of litigation carveouts in these agreements and the “California” effect on these provisions.
- The Contracts Sample and the Prevalence of Arbitration Clauses
We generated a sample of CEO employment contracts by creating a list of all of the companies included in the S&P 1500 from 1995 to 2005. Using this list, we examined each of these companies’ mandatory securities-law filings under the 1934 Securities Exchange Act in the SEC’s EDGAR database. We employed a privately owned version of this database, Live Edgar, for ease of manual and electronic search techniques.
Using Live Edgar, we checked each company’s Form 10-K (annual report) filings to see if the company mentioned that its CEO had an employment contract. If so, we searched for these employment contracts, which are attached as an exhibit to one of the firm’s securities filings. If they exist, these contracts are required disclosures for every registered company. Whenever we found one of the contracts, we downloaded it and coded it using a coding system that we created in order to gather the requisite information. We found a total of 1970 contracts in this search.
We found several different variations in CEO employment contract types. For our purposes, the three most important contract types are initial contracts, contract amendments, and restated contracts. Initial contracts are those that are entered into between the company and its new CEO, or in some cases, by a company and its current CEO when the firm’s prior relationship with the CEO was not the subject of a written employment contract. Generally, CEOs and firms enter into initial contracts at the beginning of their employment relationships.
Contract amendments can be initiated at any time for any reason once an employment contract is in place. They are typically quite short and affect only a few terms of the initial (or restated) contract, usually specifying changes in the CEO’s compensation arrangement. They rarely alter any noncompensation-related terms of the employment relationship. We did not find a single case where these amendments altered the arrangements, or lack thereof, providing for arbitration of any dispute between the parties. For this reason, we decided to drop these contract amendments from our sample. This decision reduced our sample by 1052 observations.
Restated contracts (sometimes called “amended and restated contracts”) are contracts that are entered into subsequent to the initial contract, usually after one or more amendments have been made to the initial contract. A restated contract incorporates all of the changes made in the various amendments, and it also frequently adds new terms. This new, integrated document reflects all of the terms of the employment contract between the CEO and the firm. We included these agreements in our sample for three reasons. First, in some instances, the initial contracts are unavailable. Second, restated contracts can result in changes to the arbitration arrangements in the parties’ agreement. Finally, they provide us with some insight into the “stickiness” of arbitration provisions, that is, whether their presence in an earlier CEO contract is likely to persist in later ones.
Next, for some of the variables included in our multivariate analysis, we needed additional information about the CEOs and their companies. For this information, we used the Compustat database maintained by Standard and Poor’s. It contains company-specific descriptive and financial-statement data for the past twenty years on an annual basis for companies that have actively traded securities at any point during that time period. The Compustat database enabled us to collect information about firm performance, the company’s principal business line, firm size, and the state of the headquarters for the firms where our CEOs worked. The firm performance variables are sales and return on capital. Return on capital is defined as net income divided by capital. Capital is the book value of debt and equity as reported on the firm’s balance sheet. Return on capital is then averaged over the five-year period prior to the start date of the CEO’s compensation contract. After eliminating contracts for which this information was unavailable, we ended up with 910 contracts in our sample.
For each employment contract in our sample, we coded the presence or absence of an arbitration clause. For each of the employment contracts containing an arbitration clause, we coded a wide variety of additional information about the contents of the arbitration provision, including whether the parties: (1) chose an arbitration association (and if so, which one); (2) specified the rules of procedure that arbitration would follow; (3) chose a specific number of arbitrators to resolve disputes; (4) chose a location for arbitration (and if so, where); (5) contractually allocated the costs of arbitration; (6) provided a limitations period for filing for arbitration; (7) addressed rules of discovery or rules of evidence;
- contractually prohibited the arbitrator from awarding punitive damages;
- were bound to keep the contents of arbitration proceedings
confidential; (10) obligated the arbitrator to issue a written opinion justifying her decision; and (11) reserved the right to appeal the arbitrator’s decision to an arbitration-appeal panel or to the courts. In addition, we coded the contracts for whether the parties carved out a right to have courts grant preliminary relief and whether the parties carved out particular types of claims for court resolution, including claims involving the confidentiality, noncompete, nonsolicitation, and nondisparagement provisions of the agreement.
Turning to the presence or absence of arbitration provisions, we see in Table 1 that of the 910 contracts studied, 469 (52%) provided for arbitration of at least some of the parties’ disputes, and their arbitration provisions typically required both parties to proceed to mandatory, binding arbitration. This proportion is consistent with the results of two earlier studies by one or more of the authors of this Article.
In an earlier study by the authors of this Article, we noted that the overall average of about half masks an upward trend in the use of arbitration over time—from an observed low of 36% of the contracts in 1997 to a high of 60% of contracts in 2005. A simple linear regression of the percentage of contracts containing arbitration provisions each year against time indicated a statistically significant upward slope. This trend indicates a possible transition toward a new equilibrium situation where arbitration is more commonly employed than in the past.
|Presence of Arbitration Clause in Contract by Location and Firm Type
As discussed in Part II, we had good reasons to believe that firms located in California might be more likely to insert arbitration provisions into contracts with their employees. To determine whether California law is influencing contracting parties, we separated firms based on the designated primary location of the firm in Com pus tat. Primary location is likely a rough, albeit imperfect, proxy for the location of employment of the CEO. However, even if it does not capture the actual headquarters of a firm, a firm primarily located in California should be concerned about the possible application of California law to the agreement.18‘ In our sample, 112 contracts were for firms that had their primary location in California, while the remaining 798 were for firms located primarily in other states. Of the 469 contracts that contained an arbitration clause, 75 of those contracts involved firms primarily located in California (“California firms”), and 394 involved firms primarily located outside of California (“non-California firms”).
We tested a null hypothesis that the incorporation of an arbitration clause into an agreement did not depend on whether the firm was primarily located in California. As Table 1 shows, 67% of contracts entered into by California firms contained an arbitration clause, compared to 49% of non- 
California firms. Using a chi-squared test for statistical significance, we reject the null hypothesis of no difference. This suggests that California firms are more wary of courts in general than are non-California firms.
We also wanted to know whether information technology firms (“IT firms”) are more likely to want to arbitrate disputes than other types of firms (“non-IT firms”). We had no clear hypothesis regarding the question, however. On the one hand, if IT firms can garner more value from noncompete clauses that are reliably enforced across state borders, then we would expect IT firms to choose arbitration more often than non-IT firms, given that some courts will heavily scrutinize noncompete clauses in employment contracts and California courts refuse to enforce them altogether. On the other hand, a cursory glance at our collected data indicated that companies commonly carve out noncompete clauses and other information-protecting clauses for court resolution. If IT firms disproportionately rely on such provisions to protect firm value, they might be inclined to forgo arbitration altogether in order to ensure court protection of the rights they most value. To test whether there might be a difference in the use of arbitration clauses for IT firms, our null hypothesis was that there is no difference. To test this claim, we coded the firms into IT firms and non-IT firms. The firms were categorized according to whether they were classified as being in the information technology economic sector using the Global Industry Classification Standard (“GICS”) as reported in Compustat. Of the CEO contracts having arbitration clauses, there were 115 CEO contracts for IT firms and 354 CEO contracts for non-IT firms. Table 1 shows that 62% of IT firms had arbitration clauses in their contracts, whereas only 49% of non-IT firms did. The difference is statistically significant, indicating that the IT firms have a definite preference for arbitration.
- Arbitration Procedure and Cost findings
We turn next to the contents of the arbitration provisions. We began by examining whether or not the parties chose a specific arbitration association. In our sample, when the parties chose arbitration, they almost always specified an arbitration association that would handle their disputes, and they also virtually always chose a governing set of procedural rules for arbitral proceedings. As shown in Table 2, in 435 out of 469 (93%) of the contracts with arbitration clauses, the parties chose an arbitration association, and in another 6 contracts specifically selected a method for arbitration without choosing a specific association (for a total of 441 contracts). Only 28, or about 6%, of the arbitration clauses in our sample remained silent on the method of arbitration. Similarly, Table 3 shows that 96% (all but 18) specified a set of governing rules to apply in arbitration.
While the parties almost always chose an arbitration association and a set of governing rules, there was remarkably strong convergence on one of two arbitration associations but considerable variation across the choice of arbitration rules. Table 2 shows that the overwhelming majority of contracts choosing an arbitration association chose either the AAA (399 out of 441, or 90%) or JAMS/Endispute (30 out of 441, or 7%). Seven other arbitration associations were chosen, although only in one or two contracts each in our sample.
Party Choice of Arbitration Association
* JAMS and Endispute merged into JAMS/Endispute in 1994, so these choices are actually identical.
Turning to Table 3, we show the parties’ choice of governing arbitration rules. Overall, some version of the AAA rules of procedure were chosen in 306 of the 451 contracts (68%) specifying rules for the arbitration. Although substantial, the percentage of parties choosing AAA rules is a much smaller percentage than the percentage of contracts choosing AAA as a forum for dispute resolution.1^ This disparity suggests that many contracting parties chose arbitration associations for reasons  other than the desirability of their rules of procedure. In all, seventeen different types of arbitration rules were chosen. Given the apparent textured choices made, it appears that the parlies gave this feature more attention than they afforded to the choice of arbitration association. Future empirical studies should focus on the overlap between arbitration associations and arbitration rules.
|Party Choice of Arbitration Rules
Table 4 shows that the parties also commonly addressed the allocation of arbitration costs. Overall, 355 out of the total of 469 contracts with arbitration clauses, or 76%, state how the parties intended to allocate costs, although 114, or 24%, remained silent on the issue. These numbers are somewhat misleading because the choices do not all contribute to the degree of customization. For example, 47 of the contracts left the allocation of costs to be determined by the arbitrator, a provision that is no more informative than silence. Another 52 of the contract provisions stated either that each party will bear its own costs, or that the costs will be split evenly
between the parties, which are the two most likely outcomes when cost allocations are determined by the arbitrator.’86 Even so, 252 out of 469, or 54%, of the arbitration clauses clearly customized the allocation of arbitration costs in that they displaced the applicable legal default rule. These latter provisions typically specified either that the employer would pay a majority of the costs of arbitration (90 contracts, or 19% of all arbitration clauses), or that the costs would be allocated based on the resolution of the dispute (162 contracts, or 35% of all arbitration clauses), a type of prevailing-party provision.
Party Choice of Allocation of Arbitration Costs
Our contracts provide useful insight into a relative preference for the English rule relative to the American rule for awarding attorney fees. In a recent study, Ted Eisenberg and Geoff Miller examine a broad variety of contracts filed with the SEC to glean a relative preference for each rule for the allocation of attorney fees among sophisticated parties.l87 Under the American rule, each party bears its own attorney fees, but under the English rule, the loser pays the attorney fees of the prevailing party.’88 Eisenberg and Miller categorize a silent contract as one that chooses the American rule, even though silence could simply mean that the parties have not considered   
the issue. Moreover, many of their contracts contain arbitration clauses, and silence in an arbitration clause does not necessarily indicate that the American rule would apply by default. Given that the allocation of attorney fees is most often left to the arbitrator to decide, silence in the context of a contract with an arbitration clause is much more likely to indicate no preference regarding the split of attorney fees. With silent contracts categorized as a preference for the American rule, Eisenberg and Miller find no preference between the American rule and the English rule: 39% of contracts fall in the former group and 35% of contracts fall in the latter group. Because we have good reason to believe that our silent contracts most likely indicate no preference, we can directly compare the number of contracts choosing the American rule, at most 52, with those choosing the English rule, 162. In our contracts, where parties expressed a clear preference, they were more than three times as likely to choose the English rule than the American rule. These contrary results are particularly striking, given that Eisenberg and Miller hypothesize that relational contracts (including employment contracts) should be more likely to choose the American rule.
- Other FEa tures of Arbitration Customiza tion
Beyond these three factors—arbitration association, governing rules, and cost allocation—relatively few contracts customized the features of future arbitrations. This is clearly illustrated by the data presented in Table 5. In our contracts, 57, or 12%, of the arbitration clauses included a provision that required the parties to keep confidential the details of a dispute. Only 47, or 10%, of the arbitration provisions stated that the arbitrator was required to provide a written statement of her findings. Roughly the same number of clauses prohibited the arbitrator from awarding punitive damages. In 30, or 6%, of the arbitration clauses, the parties agreed to permit discovery and specified how it would be conducted. We also find that in 21, or 4%, of the clauses, the parties limited the period of time after a claim arose in which parties could file an arbitration request. Even more rarely, in 8, or 2%, of the clauses, the parties agreed about what was permitted testimony in the arbitration hearing. And finally, in only 3, or 0.6%, of the clauses, there were provisions addressing the parties’ right to appeal an arbitration decision to courts.
Customization of Contracts by Feature
We find this lack of customization to be surprising given the amount of academic attention that has been devoted to the issue. It suggests that, at least in the heavily negotiated documents that we are examining, customization is not widespread and has been oversold in the literature.
- Carveouts: Customizing the Choice Between Utiga tion and AimrrRA tion
In contrast to the relatively low rates of customizing the specific features of arbitration, the parties more frequently specified types of claims and relief that can or must be sought in courts. Put differently, the parties often limited the scope of their chosen arbitration by carving out particular types of disputes or particular types of relief (i.e., preliminary, injunctive, or other equitable relief) for resolution by courts.
Turning to Table 6, we found that of the 469 contracts containing arbitration clauses, nearly half (219, or 47%) permitted the parties to bring at least one matter to a court for resolution. The contracts were coded for the presence of specific types of carveout provisions, including those carving out disputes involving the contract’s noncompete, confidentiality, employee solicitation, client solicitation, and nondisparagement clauses, as well as provisions providing the parties with a right to seek preliminary relief (including injunctions or specific performance) in courts.
Confidentiality carveouts permit the parties to bring to court disputes involving any provision of the contract addressed to confidentiality, the nondisclosure of information, or the protection of trade secrets. Overall, 37% of the contracts with arbitration clauses and 79% of contracts that had at least one carveout specified that disputes involving the confidentiality clauses would or could be resolved in courts.
Noncompetition carveouts allow the parties to remove from arbitration disputes that involve the noncompete clause (or covenant not to compete) contained in the employment contract. In all, 32% of the contracts with arbitration clauses and 69% of contracts containing at least one carveout specified that disputes involving these noncompete clauses would or could be resolved in courts.
Employee-solicitation carveouts allow disputes involving a contract provision that forbids the executive to entice an employee to leave the firm for employment elsewhere to be litigated in court. We found this carveout in 30% of the contracts with arbitration clauses and 64% of contracts containing at least one carveout. Less common carveouts we found were: client-solicitation carveouts1^ (22% of all contracts with an arbitration clause); disparagement carveouts   (7% of contracts with an arbitration clause); and preliminary relief carveouts,95 (12% of contracts containing arbitration clauses).
Number of Contracts with Arbitration Clauses Containing Carveouts
by Type of Claim Carved Out
|Type of Carveout||Percentage of Total Contracts with This Carveout
(N = 469)
|Percentage of Contracts
|Percentage of Contracts at NonInformation Technolog)’ Firms
(N = 354)
|Percentage of Contracts at California Firms
(N = 75)
|Percentage of Contracts at Non- California Firms
(N = 394)
Although these findings indicate that the parties often carved out particular types of disputes for court resolution, they underestimate the extent to which parties actually carved out the particular disputes—for many of the carveouts we do not know how often the contract clause on which the carveout was based actually appeared in a contract. To get an idea of how important this effect is, we coded for the presence of two very common (and therefore anticipated) types of contract provisions: confidentiality clauses and nonsolicitation provisions. In the case of confidentiality clauses, we coded contracts for the presence of any provisions requiring the CEO to keep information about the firm confidential or providing that the CEO could not disclose certain types of information. We found that 418 of the 469 contracts containing arbitration clauses also contained confidentiality provisions. Of the 418 contracts with confidentiality provisions, 172, or 41%, of them carved out these disputes for court resolution. This is somewhat greater than the 37% (with confidentiality carveouts) of all the contracts with arbitration clauses (shown in Table 6).
In the case of nonsolicitation clauses, we coded contracts for the presence of any clause prohibiting the executive from soliciting clients, customers, suppliers, or employees. We also coded contracts for the presence of any nonsolicitation carveouts, and found that 345, or 74%, of these contracts with arbitration clauses contained a nonsolicitation provision. Breaking this down further, we determined that 145, or 42%, of these contracts with arbitration provisions carved out disputes involving one or more of the nonsolicitation provisions. By comparison, Table 6 shows that employee nonsolicitation carveout clauses were found in 30%, and client nonsolicitation carveouts were in 22% of all contracts with arbitration clauses. Here too, the frequency with which disputes involving these clauses were carved out is actually significantly higher than our initial numbers indicated.
Based on our analysis in Part II, we wanted to know whether there was a difference in the likelihood that IT and non-IT firms would carve out disputes from arbitration. We hypothesized that IT firms would be more likely to include these carveouts because, relative to non-IT firms, a higher fraction of the IT firms’ value should turn on its ability to protect information and innovation. Table 6 shows our findings. Comparing these two columns of percentages, there are several apparent differences between the firm types. As we predicted, the presence of at least one carveout is greater for IT-firm than non-IT-firm contracts (52% vs. 45%), but this difference is present only for one specific category—preliminary injunctive relief (16% vs. 11%). For all other categories, non-IT-firm contracts are more likely to include litigation carveouts than are IT-firm contracts. These results surprised us initially. However, if IT firms are disproportionately located in California, the “California effect” for carveouts could be driving our IT-firm results. Our multivariate analysis in the next Part indicates that this is precisely what is happening.
Turning to the possible “California effect,” the location of the firm’s business may also have an effect on carveouts. If California law influenced the contracting parties, then we should see a difference in the use of carveouts between contracts involving CEOs to be employed at firms primarily located in California and contracts for CEOs employed at firms located elsewhere. We hypothesized that contracts involving California firms would have fewer claim carveouts because the presence of these carveouts, particularly the presence of multiple such carveouts, can cause the California courts to strike the arbitration clause altogether. Table 6 presents the data for California and non-California firms. We first tested a null hypothesis that there was no difference in the incidence of the use of the coded carveouts based on whether the firm was primarily located in California or not. The data show that when comparing the likelihood of using any carveout in an arbitration provision, non-California firms were slightly more likely to incorporate any carveout into the provisions (44% of the California firm contracts compared with 47% of the non-California firm contracts), but that difference was not statistically significant.
Regarding specific types of carveouts, however, Table 6 shows that non- California firms were much more likely to incorporate most of the types of carveouts studied. For example, 25% of non-California firm arbitration clauses carved out disputes involving the client nonsolicitation clause of the contract, as compared with only 4% of the California firm provisions. Similarly, 35% of the provisions in non-California firm contracts carved out disputes related to the employee nonsolicitation clause, as compared with just under 7% of the California firm provisions. Non-California firms were also more likely to carve out disputes involving the confidentiality and nondisparagement clauses of the agreement, and those differences were each statistically significant at the 5% confidence level.
Disputes involving the noncompetition clause of the agreement were carved out in 38% of non-California firms compared with 5% of California firms. This difference could have multiple causes, however. Given that California courts will not enforce employee noncompete provisions, firms located in California might be more likely to steer clear of asking courts to enforce these provisions even without the unconscionable-employee- arbitration-clause wrinkle.
In contrast to these carveout provisions, Table 6 shows that California firms were more likely than non-California firms to incorporate a carveout for preliminary relief. We found that 24% of the California firms’ arbitration clauses, compared with only 10% of the non-California firms’ provisions, reserved a right to seek preliminary relief in court, a difference that is statistically significant at the 1% confidence level.
These results suggest that California firms are responding to the California court precedents. The fact that California firms are just as likely to incorporate any carveouts, yet much less likely to incorporate a provision carving out any specific claim, suggests that the parties are cognizant of the California courts’ concern that multiple carveouts of firm claims creates unconscionable non-mutuality. It appears that California firms are forced to carve out fewer claims than they might otherwise carve out in order to preserve enforcement of the arbitration clause, a clause that California firms are more likely to seek in the first place.
The fact that California firms are much more likely to carve out a right to seek preliminary relief in court also suggests that they are responding to the California courts because, as mentioned earlier, the courts have made clear that simple preliminary relief carveouts do not jeopardize arbitration clause enforcement. Although the preliminary relief carveouts enable California firms to freeze assets and stop infringing behavior so that arbitrators can resolve the parties’ dispute, separation of the dispute in this manner deprives the California courts of possible efficiency gains due to the courts having jurisdiction over the entire matter.
- The California Effect: IT Firms Versus Non-IT firms
Given the surprising differences between IT firms and non-IT firms, and the differences between California and non-California firms, we investigated the possibility that there were interactive effects at work. To get a better sense of this possibility, in Table 7, we broke out the California firm contracts to compare IT-firm and non-IT-firm contracts. We found that 57% of California IT firms used carveouts, as compared with 31% of non-IT firms, a difference that is statistically significant at the 5% level. The difference initially seemed to indicate that although California firms avoid using carveouts, IT firms primarily located in California seem relatively more willing to risk enforcement of the arbitration clause in order to gain more effective protection of the firm’s information and innovation. However, when we broke down the categories and examined the differences more closely, we observed that although the IT firms were more likely than non-IT firms to utilize most forms of carveouts, few of the differences were statistically significant, and in many cases too few contracts were observed to render the results reliable. We now turn to our multivariate analysis, which provides us with a better indication of the interaction between our California effect and our results based on firm type as well as a better way to distinguish preliminary relief from claim carveouts.
Use of Any Carveout in California Firm Contracts
** indicates statistical significance at the .05 level
- Multivariate Analysis
While the univariate tests utilized in Part III are very suggestive of several important underlying relationships, we need to use multivariate regression analysis to ensure that we are controlling for the potential effects of other variables. For example, we would like to know more about how multiple factors influence the parties’ decision to place an arbitration clause into the employment contract, including the size, profitability, type, and location of the firm, and the use of an arbitration clause in prior contracts between this firm and its CEOs. In Table 8, we present the results of a multivariate logistic regression analysis using the presence or absence of an arbitration clause as our dependent variable. This form of estimation allows us to determine if the presence of any one of the independent (right hand side) variables increases or decreases the probability of an arbitration clause being included in a CEO’s employment contract. For independent variables, we include firm size, firm type (IT, non-IT), firm location (inside or outside California), an interaction term (“Cal x IT”), the presence of an arbitration clause in a prior CEO contract at this firm, and the five-year average return
on capital at the firm. These variables are defined as stated in the heading of the Table.
We are particularly interested in the California and IT variables, and the interaction between the two, in order to determine if the univariate results that we found in the earlier tables hold up. We estimate two different versions of the model. We include the variables on average return on capital, arbitration clause in prior contract, and a control variable for time, because our earlier research found that they are important influences on the presence of an arbitration clause. In Table 8, Model 1, we omit the interactive variable GA x IT, which is 1 when a firm is both California based and an information technology firm. Model 2 includes this variable.
Logistic Regression Results
Dependent variable: ARB = 1 if contract includes arbitration clause, o if not (/►VALUES LN PARENTHESES)
CALIFORNIA is a dummy variable equal to 1 if the firm’s primary location according to Compustat is in California, o if not. IT is a dummy variable equal to 1 if the firm is categorized by Compustat in the GICS information technology sector. CA x IT is an interaction variable computed as CALIFORNIA times IT. FIRM_SIZE is the In (company sales) in the start year of the CEO’s contract. INDUSTRY_CHANGE is measured by the percentage of companies in a firm’s GICS economic sector deleted from COMPUSTAT due to acquisition, merger, bankruptcy, or liquidation in the year in which the CEO employment contract started. 5- YEAR_AVG_ROC is the five-year average return on capital beginning with the year prior to the start year of the CEO’s contract. ARB_IN_PRJOR_CONTRACT is a dummy variable equal to 1 if a previous CEO contract at the company included an arbitration clause, o if not. YEAR is equal to the beginning calendar year of the contract.
|ARB IN PRIOR CONTRACT||0.27*||0.28*|
|Likelihood ratio Chi-Squared||38.21***||38.36***|
|***, **, * indicate statistical significance at the .01. .05, or .10 level, respectively|
The results show that there is a strong, consistent California effect on the presence of an arbitration clause in CEO contracts. IT firms are more likely to put arbitration clauses in their CEO contracts as well, although this effect becomes statistically insignificant when we control for the interactive effect of a California location and an IT firm. This suggests that IT and nonIT firms may be equally likely to wish to protect their information, reputation, and innovation more efficiendy with the use of arbitration carveouts. In addition, the five-year average return on capital, the presence of an arbitration clause in prior CEO contracts, and a time trend are all statistically significant and have the same signs, as we found in our earlier work.*°3
In results not tabulated, we further analyzed the degree of customization in arbitration provisions in CEO contracts. Using the same variables shown in Table 8, we ran a variety of different regressions for the presence and degree of customization (number of customized provisions contained in a single arbitration clause). We found that there was a strong, significant positive time trend in both the presence of customization and its degree. This indicates that more contracts are being customized over time and to a greater degree. How’ever, none of the other explanatory variables were statistically significant.
Can carveouts be predicted using the same variables? This is a way of testing the persistence of the univariate effects we found in Tables 6 and 7. Table 9 displays six different models: versions 1 and 2 examine the presence or absence of carveouts in arbitration clauses, 3 and 4 try to determine the causes for the number of carveouts in these clauses, and 5 and 6 use a dependent variable that includes only those provisions that carve out actual claims, excluding preliminary injunction carveouts. We use the same 
CUSTOMIZING EMPLOYMENT ARBITRATION
independent variables as in Table 8, and their definitions are given at the top of the table.
The results show a consistent, significant positive time trend for all six models. Carveouts are becoming increasingly common in our contracts, and more carveouts are appearing in each contract. In Models 3, 4, 5, and 6, we find a strong, statistically significant, and negative relationship between firms primarily located in California and the number of carveouts contained in their contracts. California firms clearly use fewer carveouts than firms located outside of California.
Model 2, which examines the presence of any carveout, offers some intriguing results as well. There is a weakly significant, negative effect on the presence of carveouts for California based firms, but this is offset for IT firms located in the state. California IT firms are significantly more likely than California non-IT firms to have at least one court carveout in their CEO’s employment contract. This is consistent with the univariate results shown in Table 7. Models 3 and 4 were included to determine whether California IT firms were more likely to include a larger number of carveouts from arbitration. The results indicate that the California IT firms are as reluctant to include multiple carveouts as are the California non-IT firms. Models 5 and 6 were included to determine whether California IT firms were more or less likely to use preliminary relief carveouts, which are upheld by California courts, rather than claim carveouts, which are more heavily scrutinized by those courts. We see no change in the main results reported above, indicating no significant differences across the different types of California firms.
 We made no attempt to include all of the other various contractual agreements that exist between CEOs and their firms. For example, we did not include change-of-control agreements in our sample, although some of these agreements include arbitration provisions. We recognized that these other forms of agreement may affect the employment relationship between the firm and its managers, but we decided to leave them for a later day in order to maintain our focus on employment contracts.
 This could happen if the initial contracts were entered into prior to May 6, 1996, when the SEC mandated that all companies file electronically on the EDGAR database, or if the company failed to disclose an earlier contract. We searched diligently to find all prior contracts to confirm wherever possible if they contained an arbitration clause.
 We could have included all of the CEO employment contracts in our analysis; however, the absence of the additional data on certain variables would have made it difficult to test different explanations for the presence or absence of arbitration provisions.
 Confidentiality provisions state that the executive promises not to disclose certain information about the firm and its activities.
 Noncompete provisions bar executives from competing with the company for a period of time after the termination of their employment position with the company. See Schwab & Thomas, supra note 28 at 254-57.
 Nonsolicitation provisions stop departing executives from soliciting employees and clients of their employer to change their allegiances and move to the new firm where the executive will be working.
 Nondispargement clauses are designed to ensure that departing executives do not make negative comments or statements about their former employer.
 Arbitration clauses can alternatively provide for optional arbitration or for nonbinding arbitration, where an arbitrator provides a recommended settlement. None of our contracts provided for nonbinding arbitration, and only a handful provided for optional arbitration.
 Schwab & Thomas, supra note 28; Thomas, O’Hara & Martin, supra note 50, at 981.
 Thomas, O’Hara & Martin, supra note 50, at 981.
 The t-statistic on the slope was 3.84. Id. at 981 n.71.
 See, e.g., In re Consol. Capital Equities Corp., 143 B.R. 80, 85 (Bankr. N.D. Tex. 1992) (applying California law to a corporation based on its principal place of business and location of a majority of its assets); DiTondo v. Meagher, 883 N.Y.S.2d 690, 695 (Sup. Ct. 2009) (noting that it is “well-settled” that for choice-of-law purposes a corporation is domiciled at its principal place of business).
 The difference is statistically significant at the 1% level.
 Gilson, supra note 69, at 607-09.
 These differences are statistically significant at the 1% level.
 Note that this disparity holds even if all 18 (less than 4%) of the contracts that failed to specify rules are contracts choosing the AAA as a forum and that the failure to choose governing procedural rules is a silent choice in favor of the application of AAA rules.
 See Chartered Inst, of Arbitrators, Practice Guideline 9: Guideline for Arbitrators Making Orders Relating to the Costs of Arbitration § 3.5 (remarking that it
is unusual in the U.S. for an arbitrator to award attorney fees), available at http://www.ciarb. org/information-and-resources/PracticeGuidelineg.pdf; John Yukio Gotanda, Awarding Casts and Attorneys’ Fees in International Commercial Arbitrations, 21 MICH. J. INT’L L. 1, 10 (1999) (same).
 Theodore Eisenberg Sc Geoffrey P. Miller, The English vs. the American Rule on Attorneys Fees: An Empirical Study of Attorney Fee Clauses in Publicly-Held Companies’ Contracts, (N.Y.U. Ctr. for Law Sc Econ., Working Paper No. 10-52, 2010), available at http://papers.ssm.com/s0l3/ papers.cfm?abstract_id= 1706054.
 Id. at 3.
 Id. at 4-5.
 We say “much more likely” rather than “certainly” in the text because it is possible that some arbitration associations impose rules on arbitrators for how to allocate costs in the face of a silent contract, in which case silence could indicate a choice in favor of the association’s rule. However, many associations do not so constrain the arbitrator.
 Eisenberg & Miller, supra note 187, at 25.
 This captures the carving out of disputes involving a provision that forbids the CEO to solicit clients or customers of the firm to take their business elsewhere.
 These clauses carve out disputes involving a provision in the contract that forbids the executive to publicly disparage the company.
 This carveout permits the parties to seek preliminary injunctive relief in court in certain circumstances.
 For example, while only 7% of the contracts with arbitration clauses carve out disputes involving the nondisparagement clause of the contract, it is likely that many, and perhaps most, of these contracts do not contain a nondisparagement provision. What we really need to know is how many contracts actually include nondisparagement clauses in order to determine how frequently such claims are carved out for court resolution. Unfortunately, we did not code for the frequency of nondisparagement clauses.
 This difference is statistically significant at the 1% confidence level.
 We find that this difference is statistically significant at the 1% confidence level.
 Again, this difference is statistically significant at the 1% confidence level.
 See supra note 147 and accompanying text.
 Gary Born counsels against the carving out of claims on the ground that it invites bifurcated dispute resolution and therefore parallel proceedings. BORN, supra note 10, at 1127— 28. Note that carving out a right to proceed to court for preliminary relief but not resolution of the claims guarantees bifurcated dispute resolution. Parties outside of California may be carving out these claims for court resolution to minimize bifurcated proceedings notwithstanding the potential risks that Born identifies.
 Thomas, O’Hara Sc Martin, supra note 50, at 997.
 Thomas et al., supra note 50.