Safety regulation and the risk of workplace accidents in Quebec.

Lanoie, Paul

I. Introduction

This paper examines the effectiveness of policies adopted by Quebec's occupational safety and health (OSH) authority, the Commission de la Sante et Securite du Travail (CSST), in reducing the incidence of workplace accidents(1) after its creation in 1980. More specifically, data at the industry level (30 industries), and covering the period preceding the creation of the CSST (the "pre-CSST" period 1974-1980) and the period following its creation (the "post-CSST" period 1981-1987), will be used to ascertain, through dummy variable shift effects, whether or not there was a downward trend in accidents after the creation of the CSST in 1980. The analysis follows the methodology proposed be Curington [12] to examine the effect of OSHA (Occupational Safety and and health Administration) regulation on workplace safety in New York State.

First, it is useful to describe the institutional context. Government intervention in occupational safety and health has gained more and more attention in North America during the last two decades. In the United States, safety regulation is under the responsibility of OSHA, created in 1970 to promote workplace safety by setting numerous new safety standards and implementing measures for the enforcement of standards (inspections of firms, fines ad prosecutions). Canadian provinces followed a similar strategy in the 1970s, while adopting additional safety-enhancing measures not present in the United States. These include the right to refuse hazardous tasks, the creation of joint worksite safety committees, the requirement of a prevention program and the right to protective reassignment. The right to refuse hazardous tasks means that a worker can refuse to execute a certain task if he or she believes it to be "abnormally" dangerous. The immediate supervisor is then asked to remedy the situation and has the onus of proving that safety has been established. Joint worksite safety committees assume the following responsibilities: obtaining and disseminating information on OSHA, identifying the sources of hazard and making recommendations on means of eliminating hazards to the employer. The size of the committees varies from province to province, but equal representation of management and workers is compulsory. A prevention program must meet the approval of the OSHA board and address the training and supervision of workers, inspections, accident investigations, personal protective equipment as well as the maintenance and disclosure of records. In Quebec, prevention programs (as well as the safety committees which implement the programs) are imposed on firms with more than 20 employees in only fifteen high risk industries. Protective reassignment gives a worker the right to be transferred to another job within the same firm if he or she can provide a medical certificate that attests the potential harm his or her job could cause. So far, this right can only be used by pregnant women. The last two measures (prevention programs and protective reassignments) have been adopted only in the province of Quebec where the OSH board has put particular emphasis on accident prevention during recent years. As an illustration, in 1989, the CSST [7] spent $7.88 (U.S.) per worker in prevention activities, while OSHA spent approximately $1.80 (U.S.) per worker in New York State.(2)

It should also be noted that, in Canada, the government plays the role of an insurer through the presence of provincial public Workers' Compensation Boards (WCBs). In most provinces, including Quebec, workers' compensation and direct regulatory controls are under the responsibility of the same agency. Firms are considered liable for workplace accidents and pay insurance premia to the agency which, in turn, compensates accident victims. Via an experience-rating mechanism, these premia are partially adjusted to reflect the firm's own claim experience. In the United States, workers' compensation services are essentially of the same nature(3), but they are mainly provided by private insurance carriers subject to government regulation.

So far, American econometric studies using aggregate data have found that OSHA regulation has had little or no impact on workplace safety [11;12;28;29]. No existing analysis, however, has considered the performance of any Canadian OSH board in improving workplace safety. Given that Canadian province have adopted safety-enhancing policies that differ from those of American states, it is of interest to determine whether or not the influence of these innovative policies on Canadian workplace safety has been more substantial than the apparently negligible effect OSHA regulation has had on American workplace safety. The exercise is particularly relevant in the province of Quebec where the CSST has adopted two policies unique in North America.

The empirical approach suggested by Curington, and adopted here to scrutinize the effect of CSST safety policies, is useful since it allows one to identify the specific industries in which the regulation has had a discernible impact. This is an interesting exercise for policy targeting. Furthermore, it is plausible that there are "industry-specific" effects since the level of CSST enforcement varies from one industry to another (7).

Apart from the data set, the present analysis differs from Curington's in three respects. First, the empirical work is based on a theoretical principal-agent framework in which the determination of the wage is considered and in which both firms and workers can influence the risk of a work-related accident. These important features are not considered by Curington whose model is based on the firm's side of the problem. The principal-agent framework is justified since it is plausible that the firm cannot observe the risk-related behavior of its workers. For instance, an employer cannot easily monitor whether a worker takes drugs or alcohol. It is also important, in a policy analysis, that both firms and workers be able to influence the risk of accident since their respective risk-related behavior could counterbalance each other (e.g.,workers becoming negligent when the firm increases its investment in safety) and thereby undermine any safety-enhancing policy. As a result, theoretical findings analogous to Curington's are obtained, but from a more general model. Second, in the empirical analysis, a more appropriate measure of workers'compensation benefits defined as the wage replacement ratio (in which the wage is determined endogeneously) is adopted, whereas Curington uses a simpler measure, not directly linked to his theoretical model, based on the maximum insurable income. Finally, this analysis examines the overall impact of CSST policies not only on the frequency of all accidents (the conventional measure of the incidence of workplace accidents), but also on the rate of accidents that have resulted in permanent disabilities. This is motivated by a potential problem of accident reporting. Indeed, as argued by other authors (25), it seems reasonable that the reporting of accidents increased over the period 1974-1987. For instance, since its creation, the CSST has opened eleven regional offices so that accident compensation can be handled locally. It is plausible that such a measure has facilitated the reporting of accidents (7). The fact that the relation between certain illnesses and certain types of jobs has become better known through time can also explain an increase in accident reporting (8). If this is the case, any ameliorating effect of the CSST on the incidence of accidents could be counterbalance be better reporting. However, it is arguable that accidents that generate a permanent disability (such as the loss of a limb) are more likely to have been reported in the same manner through time, attenuating any reporting bias. No researcher has previously examined the effect of safety-enforcing measures on a category of workplace accidents presumably non-biased with respect to accident reporting. Curington, for instance, has a different focus, concentrating on categories of accidents that have a better chance of being prevented through safety regulation (such as "struck by" or eye injuries).[4]

The test of the paper is organized as follows: Section II presents the theoretical model underlying the empirical analysis. Section III discusses the estimation technique, the data and the specification of the equations to be estimated. Section IV presents the instrumental variable estimates of the risk equations that suggest that CSST safety-enhancing measures have been successful in reducing the risk of accidents in certain industries. However, there seems to be no evidence that better reporting could have counterbalance any ameliorating of the CSST policies on accidents. Section V provides concluding remarks.

II. The Theoretical Model

The theoretical model adopted here differs from previous work in that it uses a principal-agent framework in which a firm and a worker play a Stackelberg game. As will become apparent, this setting seems to be an appropriate representation of a situation in which workplace accidents can occur (modified version of the model have been discussed and presented elsewhere [19;20]).

To focus on the "market of workplace accidents", the economy is simplified in some respects. Consider a representative risk-neutral firm operating in a competitive industry "i" at time "t" in which there is a risk of an accident. The firm, which has no capital, hires a fixed number of identical risk-averse employees each working a fixed number of hours during the period and making one unit of product. The risk of an accident depends upon the level of risk-preventing activity by the firm and the worker.

It is important to specify the timing of the problem. The events in this static model occur within one period, but their timing is not the same. The worker receives a wage, w, per period and chooses an accident-prevention effort, e, per period over a continuum of non-negative effort levels. This effort can be interpreted in terms of time spent in risk-preventing activities or the unpleasantness of these activities (e.g., wearing safety glasses). The firm selects a non-negative safety expenditure per worker of q. Examples of this kind of expenditure would be safety equipment distributed to workers or investment in safer technology.(5) These variables should be thought of a occurring continuously through the period (discounting is ignored). There may or may not be an accident. For simplicity, it is postulated that only one type of accident of fixed severity exists (Curington explains the severity of accidents in his model). The accident, if it occurs, takes place on the last day(6) of the period and does not, therefore, affect a worker's output. The worker receives an exogenously given compensation benefit [Alpha] on that day from a WCB(7). This treatment differs from that of other authors [23; 6], who implicitly assume that an accident can happen only on the first day of the period.(8) The treatment adopted here is no more extreme than theirs, and it is in intended to present, within a one-period model, the fact that receipt of wages is continuous, while accidents are rare and discrete.

The probability of an accident, [P.sub.it], is a decreasing function of e and q (i and t are suppressed where there is no ambiguity): (1) P = P (q, e); [P.sub.q], [P.sub.e] < O; [P.sub.qq] [P.sub.ee] > O. The sign of [P.sub.qe] may be positive or negative. As explained in previous s work [23], some expenditures by the firm may inform employees (signs, for example) and act as complements to their efforts ([P.sub.eq] < O). However, other precautions taken by the firm, such as safety devices on a machine, are likely to be perceived as substitutes for workers' efforts ([P.sub.eq] > 0).

The worker's expected utility, EU, depends on consumption if there is no accident (consumption is assumed to equal income), compensation in the event of an accident, effort and the probability of an accident. With appropriate simplifying assumptions, EU becomes: (2) [EU.sub.it] = P(q, e) [U.sup.a] + (1 - P(q, e)) [U.sup.n] - e. It is an event-dependent utility function where [U.sup.a] is the worker's utility function if there is an accident and [U.sup.n] is the utility function if there is none. Since w and [Alpha] are perceived at different time in the period, it is not unreasonable to postulate that [U.sup.a] is additively separable in w and [Alpha]. Therefore, the function [U.sup.a] is assumed be the sum of the worker's utility during the period (or before the realization of the uncertainty), U (w), which depends only on w, and the utility on the last day, [Z.sup.a] ([Alpha]), which depends only on [Alpha]; i.e., [U.sup.a] (w, [Alpha]) = U(w) + [Z.sup.a] [Alpha]. Similarly, [U.sup.n] (w) = U (w) + [Z.sup.n], where [Z.sup.n] is a constant representing the utility on the the last day if there is no accident . Furthermore, [U.sup.j] (*) for j = a,n is a function from R + into R with [Mathematical Expression Omitted] > 0 and [Mathematical Expression Omitted] < 0 for y = w, [Alpha]. Obviously, the focus is put on situations in which a worker is not better off because of an accident; i.e., it is assumed that [Z.sup.n] > [Z.sup.a] in equilibrium (which implies that [U.sup.n] < [U.sup.a]. Finally, since the effort is undertaken before it is known whether or not the accident occurs, the disutility of effort is assumed to be event-independent [1]. For brevity, P, [U.sup.a] and [U.sup.n] are often used without their arguments in the text.

A representative firm in industry i at time t has an expected profit function per worker: (3) [Mathematical Expression Omitted]

This function indicates, that, for each worker, the firm receives the prices of his or her output, f, while it pays a wage, w and spends an amount of safety expenditures per workers, q. Furthermore, the firm faces an expected cost for not complying with safety-enhancing policies promulgated by the OSH authorities. These policies can be modelled by considering a term G times (q [bar] - q) in the expected profit function where G is the expected cost to the firm for not complying with safety regulations requiring an amount of safety expenditure q [bar] (9).

Furthermore, firms pay an insurance premium to a WCB and, for simplicity, it is assumed that experience rating is perfect.(10) When the WCB makes no profit(11) and provides fair insurance, firms pay an insurance premium per worker equal to P [Alpha].

As argued in the [17; 19], the Stackelberg equilibrium, in which one party takes the other's reaction into account, is of interest. The interaction between firm and workers is modelled as a two-stage game where the safety expenditure, q, and the wage, w, are determined by the firm in the first stage of the game, while workers choose their effort, e, in the second stage given q and w. When the firm decides upon q and w, it has to take the workers' interest into account by providing them with a level of utility U [bar] comparable to what could be obtained in other industries. In other words, the modeling assumptions imply that q and w are enforceable in a contract between firms and workers, while e is left out. These assumptions seem realistic because, in many circumstances, a firm's expenditures on safety precede employment (e.g., the design of a machine) and can be considered as "visible" capital. A worker's effort, however, is at best imperfectly observable. Indeed, a firm may have difficulties monitoring a worker's risk-related behavior, such as running on a wet floor, although it can certainly influence this behavior. Hence, it is natural to think of the firm as committing itself by being the first actor (the Stackelberg leader) and to think of the workers as reacting (for instance, workers use the safety equipment put at their disposal).

As in dynamic programming, one first examines the second stage of the game for arbitrary levels of w and q. In this second stage, workers maximize their expected utility (2) with respect to e. The first-order condition is given by: (4) [P.sub.e] ([U.sup.a] - [U.sup.n]) - 1 = 0. In the first stage, the employer maximizes his expected profit subject to two constraints: (i) the worker's expected utility is fixed at U [bar], and (ii) equation (4) is satisfied (i.e., the solution is on the worker's reaction function).(12) The Lagrangian is: (5) [Mathematical Expression Omitted] The first-order conditions are: (6) [Mathematical Expression Omitted] (7) [Mathematical Expression Omitted] (8) [Mathematical Expression Omitted] (9) [Mathematical Expression Omitted] (10) [Mathematical Expression Omitted]

Total differentiation of equation (6) to (10), combined with P = P (e,q), leads to the following comparative-static results describing the "total" effect of a change in predetermined variables (government policies) on the incidence of workplace accidents, P:(13) (11) [Mathematical Expression Omitted]

The first result indicates that, as expected, a policy shift which increases the firm's expected cost of not respecting safety regulation, such as increasing fines for non-compliance with standards, leads to an unambiguous reduction in the incidence of workplace accidents. The second result, also not surprising, shows that an increase in compensation benefits [Alpha] has an ambiguous impact on P. Indeed, a rise in [Alpha] reduces the opportunity cost of an accident for workers, inducing them to be less careful; whereas it raises the opportunity cost of an accident for the employer (given perfect rating), inducing the firm to devote more resources to safety. Therefore, [dP.sub.it] / [d [[Alpha].sub.it]] < 0 if employee responses responses, whereas [dP.sub.it] / [d [[Alpha].sub.it]] > 0 if the converse occurs. Overall, this model generates a theoretical prediction concerning the impact of regulation that is similar to Curington's. However, the present model is more general than Curington's since it takes into account, in a realistic fashion, the behavior of the worker. Furthermore, the effect of a change in compensation benefits is considered explicit in this model, while it is not introduced in Curington's.

The sample of data covers both the period in which there was little or no government intervention, and the period following the creation of the CSST in which extensive safety-enhancing policies were implemented.(14) With such a sample, it is not possible to include-enhancing measures as independent variables in a regression analysis to verify explicitly whether the comparative-static result, [dP.sub.it] / [dG.sub.it] < 0, holds. Given these circumstances, the overall negative impact of CSST policies on the incidence of accidents should be captured as a significant downward trend or shift occurring after the creation of the CSST.(15) The downward trend could be perceptible in each industry through the appropriate use of industry dummy variables as in Curington[12].

III. Data, Specification of the Frequency Equation and Estimation Technique

For the estimations, pooled cross-section and time-series data from Quebec is used. The data are on a yearly basis from 1974 to 1987 (14 years) and at the two and three-digit industry level (30 industries covering most sectors of the economic activity, see Table I for the list of industries). The nature of this sample is similar to Curington's which includes 18 manufacturing industries from New York State for the period 1964-1976 (13 years). In particular, in both samples, the periods preceding and following the policy change have approximately the same length (OSHA was created in 1970). Table I provides the exact definition of all the variables used in the analysis, their mean, standard deviation and statistical source. [Tabular Data I Omitted]

In the empirical work below, the function P is assumed to be linear and an equation of the following form is estimated (this formulation is inspired by Viscusi [29]): (12) [Mathematical Expression Omitted] It is assumed that the equation explaining the rate of accidents with permanent disabilities ([PERMRATE.sub.it]) has the same specification.

[FREQUENCY.sub.it] is the frequency of workplace accidents in industry i at time t defined, as [PERMRATE.sub.it], under its log-odds form so that the transformed frequency rate is not constrained to the (0, 1) interval. DUMCSST and [INDUSPOST.sub.i] will be described below in the estimation technique used by Curington. As in Curington, as set of industry dummies ([INDUSTRY.sub.i]) for the whole sample is included to control for the fact that the level of inherent risk may vary from one industry to another. [X.sub.kit] is a vector of control variables and [e.sub.it] is the error term. As argued elsewhere [30], the generosity of compensation benefits ([Alpha]) is measured by a variable ([COMP.sub.it]) expressed in logs (In) based on the net wage replacement ratio obtained by a disabled worker in the case of a temporary total disability (taking into account the maximum insurable income, see Table I). Furthermore, following previous analyses [28;29], a lagged dependent variable is included to serve as a proxy for the safety conditions that prevailed during the previous period. Therefore, the current of accidents adjust to changes in dependent variables with a geometrically decreasing lag structure.

The control variables included in the vector [X.sub.kit] are socio-economic variables that were shown to be relevant in the literature. It is postulated that, ceteris paribus, industries with a higher percentage of female workers ([FEMA.sub.it]), older workers ([AGE45.sub.it]) and relatively well-educated workers ([EDUC.sub.it]) will fewer accidents. Indeed, it is expected that jobs with a higher fraction of educated and female workers should involves less physical effort and pose lower risk [17; 29]. Older workers, because they are usually more experienced, should have fewer accidents [12]. Alternatively, the higher the percentage of younger workers ([AGE24.sub.it]), less educated workers ([EDUC1.sub.it]) and workers belonging to ethnic minorities ([MINOR.sub.it]),(16) the greater the incidence of accidents should be. The coefficient of the percentage of unionized workers ([UNION.sub.it]) may take either sign. If unionized workers have a greater propensity to report accidents than others, [UNION.sub.it] can take a positive sign [5]. However, if unionized workers are more informed and conscious of accident prevention, one might expect a negative sign on [UNION.sub.it] [12]. The proportion of married workers ([MARR.sub.it]) may also take either sign. Married workers may have an incentive to be more careful due to their additional responsibilities, although any revenue from a spouse can be considered as an additional form of insurance that could lead married workers to be less careful. The ratio of machinery and equipment and equipment to labor ([MACHLAB.sub.it]) is likely to take a positive sign because a worker's risk usually increases according to his or her contact with machinery [12]. The unemployment rate per industry ([UNEMP.sub.it]) is included to capture business cycle effects and is expected to take a negative sign [29]. Indeed, during a cyclical upswing, the price of foregone output increases and therefore, the cost of devoting inputs to prevention rather than output rises. This should induce firms to reduce their safety expenditures which should lead to more accidents. Moreover, if the work pace accelerates during a cyclical upswing, the risk of accidents increases.

Estimation Technique

Since the CSST was created in 1980, the dummy variable DUMCSST is used, which is equal to one for each observation in the years 1981-1987 and zero otherwise. [INDUSPOST.sub.i] is a vector of industry dummies equal to one for each industry i for the years 1981 to 1987 inclusively, and zero otherwise. In this context, DUMCSST represents the pre-post difference in the accident rate for the reference group (the default industry), and each variable in the vector [INDUSPOST.sub.i] represents the difference between the change in the accident rate in the reference group and the change in the accident rate for industry i. Therefore, the sum of the coefficient of DUMCSST and the coefficient of the variable in the vector [INDUSPOST.sub.i] that corresponds to industry i is an estimate of the post-CSST change in the accident rate for industry i. One can determine whether that change is statistically different from zero in industry i. The standard error used to estimate the required t-ratio is the square root of the estimated variance. (17)

In the econometric analysis, the instrumental variance (IV) technique is used since the net wage replacement ratio from the CSST (COMP) is a function of the wage, w, which is an endogenous variable in the theoretical model. To obtain an instrumental variable for COMP, predicted values are computed for this variable. The latter are obtained by first regressing the wage rate on all the exogenous variables (including the wage at time t -- 1 which serves as an instrument) and then replacing the wage rate by its predicted value in the definition of COMP.[18] Curington uses a simpler measure of compensation benefits, defined as the maximum insurable income divided by the industry wage.

Furthermore, since the sample consists of observations on industries that vary greatly in size, there is a reason to suspect conditional heteroscedasticity. A series of Breusch-Pagan [4] test were performed by regressing the square-residuals of the estimated equations on the total employment of each industry; the sample size times the [R.sup.2] of these regressions asymptotically follows a chi-square distribution with one degree of freedom. The results of these tests (available on request) showed no evidence of conditional heteroscedasticity for the FREQUENCY equation. However, there are some evidence for the PERMRATE equation and the IV analogs of the weighted least-squares technique are used for that equation [3,90-96].

A Lagrange Multiplier test [14] was also performed to detect first-order serial correlation for each industry. For this purpose, different autocorrelation coefficients (p) were allowed across industries. The tests showed no evidence of first-order serial correlation.(19)

Overall, one has to keep in mind the limitations of the exercise since a post-CSST change in the accident rate could be due to factors other than safety-enhancing policies that are not captured in the analysis. For instance, it is noteworthy that the period immediately following the creation of the CSST (1981-1983) was characterized by an important recession the effect of which, however, should be captured by the variable UNEMP which reflects cyclical variations in economic activity.(20)

IV. Empirical Results

Table III reports statistically significant post-CSST changes in the accident and permanent disability rates. These results were shown to be robust to the exclusion of the lagged dependent variable or the use of the variable COMP under its linear rather than logarithmic form (complete results are available upon request). Table III shows that there was a significant decline in the frequency of all accidents in four industries: construction (-4.7%), manufacturing of transportation equipment (-5%), manufacturing of electrical products (-1.1%) and miscellaneous manufacturing industries (-3.9%). However, there was a significant decline in the permanent disability rate in only two industries: miscellaneous manufacturing industries (-1.5%) and trade (-1.6%), while there was a significant increase in the rate of permanent disabilities in the hosiery and apparel industry (+1.4%). Therefore, there is no clear evidence that ameliorating effects of the CSST on the incidence of all accidents were counterbalanced by better reporting. Were this the case, one would have observed greater declines in the rate of permanent disabilities (presumably non-biased with respect to accident reporting)(21) than in the overall accident rates after the creation of the CSST. However, it is not entirely certain that better reporting not counterbalance any ameliorating impact of the CSST. Indeed, these results could simply mean that there is better reporting, but CSST safety-enhancing policies are more efficient in preventing minor injuries than more severe ones. Safety-enhancing policies may have no effect on severe accidents if they occur because of unpredictable circumstances, or if they occur when workers temporarily do not comply with the regulation. Given the available data, it is not possible distinguish between reporting and prevention effects.

Furthermore, it is noteworthy that in Curington's Table III, many industries are shown to have experienced an increase in the frequency of all injuries after the creation of OSHA, while only one experienced a significant decline in its overall accident rate (6.5% in the fabricated metal products). Therefore, the results of this study, which shows a better performance in terms of accident prevention, may suggest that the innovative policies adopted by the CSST to promote safety (such as the right of refusal or safety committees) are more efficient in reducing the incidence of accidents than those prevailing in New York State. However, these results could also mean that conventional safety policies (inspections, fines) are more strictly implemented in Quebec than in New York State. [Tabular Data III Omitted]

Concerning the other variables in the equations, Table II shows that the coefficient of the variable reflecting the generosity of compensation benefits (ln COMP) is negative in both the FREQUENCY and the PERMRATE equations, although it is non-significant. This result is different from what is observed in Curington (Table I, first column), and this could be due to the definition of the compensation variable used in this text. Furthermore, as in Curington, the coefficient of the ration of machinery to labor ( MACHLAB) and of the proportion of workers to ethnic minorities (MINOR) is shown to be positively associated with the frequency of all accidents, the percentage of married workers (MARR) is negatively associated with the permanent disability rate, while the coefficient of the percentage of younger workers (AGE24) is negative and significant in the PERMRATE equation. This last result indicates that, although younger (less experienced) workers may have more minor accidents than other workers (as suggested in the rest of the literature), they are better able to avoid major accidents (a similar result is presented in another paper [13]). Finally, the lagged dependent variable is positive and significant in both equations.

V. Conclusion

Following a methodology proposed by Curington [12], this paper has examined the "overall" effectiveness of policies adopted by Quebec's OSH authority (CSST) in 1980 to promote safety in the workplace. In particular, the analysis is innovative in that it refers to a principal-agent theoretical framework, it uses a more precise definition of compensation benefits and it considers the impact of safety-enhancing policies on a category of accidents presumably non-biased with respect to accident reporting: the permanent disability cases. The results suggested that CSST safety-enforcing measures were able to reduce the incidence of all accidents and of permanent disability cases in certain industries. However, there seems to be no evidence that better reporting could have counterbalanced any ameliorating impact of CSST policies on accidents. (1)Throughout the text, an accident may be interpreted as an injury or a disease. (2)The American figure was provided by an official of the OSHA Regional Office for Region 2 (this Region covers New York, New Jersey, and Puerto Rico). (3)As an illustration, in 1991, a worker in Quebec who suffers a temporary total disability (these cases represent approximately 85% of all the compensable accidents in Quebec) receives 90% of his or her net wage (non-taxable) subject to a maximum insurable income of $702 U.S./week. In New York State, a worker with the same type of disability receives 2/3 of his or her gross income (non-taxable) subject to a maximum insurable income of $540 U.S./week. (4)Data on specific types of injury is not available in Quebec before 1980. (5)It is plausible that precautions could affect productivity. Nevertheless, this possibility is ruled out since, as the reader can verify, this would add complexity to the model without introducing any new basic insights. (6)The formulation adopted here obviously involves collapsing a multi-period problem into a single-period analysis; given the purpose of the text, there is no loss in doing so [2, 4]. (7)For reasons not specifically modelled here, such as adverse selection, it is assumed that workers cannot insure themselves in a private market. (8)Another author [18] adopts a similar accident timing in his two-period model in which an accident can happen only in the second period, and the disabled workers produce no output in the second period. (9)G is actually the product of the probability of the firm being caught in non-compliance and the infraction penalty. There is no reward from the government for q > q [Bar] [28] and this case is ignored. (10)Perfect experience rating implies that the insurance premium paid by the firm is equal to the cost compensating disabled workers within the firm. The reader can verify that the nature of the results is not altered if experience rating is assured to be imperfect. (11)WCBs are public enterprises in Canada and regulated firms in the United States and, therefore, are assumed to be non-profit. (12)See Jewitt [16] and Rogerson [24] for a discussion on the validity of this approach. (13)The computation of these results (which hold whether [Mathematical Expression Omitted] and with the assumption that third-order terms are equal to zero) involves straightforward algebra and can be found in a previous discussion paper [21], Appendix I. (14)In the period preceding 1980, there was a board in Quebec--the "Commission des Accidents du Travail"--which played the role of insurer now played by the CSST. There were also safety inspections performed by three Ministries (Labor, Mining and Environment), and by a board related to the construction industry (Office de la Construction du Quebec or OCQ). However, it was not possible to obtain reliable information on these inspections. For instances, OCQ inspectors also checked the quality of the buildings constructed, so that it is not clear which proportion of the inspections reported by the office was aimed at ensuring workers' safety. (15)Data concerning the compensation [Alpha] is available for her whole sample, so that it will be possible to determine directly whether [Mathematical Expression Omitted]. It is noteworthy that, in 1979, there was a change in the Quebec compensation regime. The change was intended to provide greater compensation for accident victims with gross annual income less than $9000.00 (Can.) at the expensive of those whose gross annual income exceeded this limit. This change will be captured in the available COMP (to be defined below). (16)A plausible explanation for the higher rate of accidents among ethnic minority is potential language problem at the level of understanding safety orders. (17)The exact formulation of the standard error is: [Mathematical Expression Omitted] where

S= estimated standard error. [B [caret].sub.1] = estimated coefficient of DUMSCSST. [B [caret].sub.i] = estimated coefficient of the variable in the vector [INDUPOST.sub.i] that

 corresponds to industry i.
EST. COV = estimated covariance.

(18)Strictly speaking, this means that a "semi reduced-form" equation is estimated. (19)Tests were also performed to detect second-order serial correlation, but no evidence of this was found either. (20)One should also note that there are no time dummies in the model so that the results can easily be compared with Curington's. The implicit assumption behind this specification is that the most important time-specific effect is the introduction of the regulatory board, which is captured through DUMCSST. However, time-specific effects other than the introduction of regulation and cyclical variations are plausible. For instance, one could argue for a trend in the accident rate (not considered by Curington) due to changes in technology [25]. To investigate this potential effect, a linear trend was introduced (the variable TREND equal to 1 for each observation in the first year, 2 for the second year etc.). The results (available on request) were very similar to those presented here. In particular, the variable TREND was not significant in either the FREQUENCY or the PERMRATE equation. (21)As suggested by a referee, it is possible that, given the income replacement effect, permanent disabilities may be viewed as more elastic.

References

[1]Arnott, Richard and Joseph E. Stiglitz. "Equilibrium in Competitive Insurance Markets. The Welfare Economics of Moral Hazard, I: Basic Analytic." Discussion Paper No. 465, Kingston, Ont.: Queen's University, Institute for Economic Research, 1982. [2]-- and --, "Moral Hazard and Optimal Commodity Taxation." Journal of Public Economics, February 1986, 1-24. [3]Bowden, Roger J. and Darrell A. Turkington. Instrument Variables. New York: Cambridge University Press, 1984. [4]Breusch, Trevors and Adrian Pagan, "A Simple Test for Heteroscedasticity and Random Coefficient Variation." Econometrica, September 1979, 1287-94. [5]Butler, Richard J. and John D. Worrall, "Workers' Compensation: Benefit and Injury Claim Rates in the Seventies." Review of Economics and Statistics, November 1983, 580-89. [6]Carmichael, Lorne, "Reputations for Safety: Market Performance and Policy Remedies." Journal of Labor Economics, October 1986, 458-72. [7]Commission de la Sante et Securite du Travail. "Annual Report 1981 to 1989." Montreal: Government of Quebec, DC 400-290. [8]--. "Le regime quebecois de sante et securite au travail. Un portrait statistique et financier." Quebec: Government of Quebec, 1987. [9]--. "Statistiques sur les lesions professionnelles 1980-1984, 1979-1983, 1978-1982, 1974-1976." Montreal: Government of Quebec. [10]--. "Statistiques selon le secteur d'activite economique prioritaire." Unpublished report, Quebec: Government of Quebec, 1986. [11]Curington, William P., "Federal vs. State Regulation: The Early Years of OSHA." Social Science Quarterly, June 1988, 341-60. [12]--, "Safety and Workplace Injuries." Southern Economic Journal, July 1986, 51-72. [13]Dillingham, Allan E., "The Effect of Labor Force Age Distribution on Workers' Compensation Costs." Journal of Risk and Insurance, June 1982, 235-49. [14]Godfrey, Leslie G., "Testing for Higher-Order Serial Correlation in Regression Equations When Regressors Included Lagged Dependent Variables." Econometrica, November 1978, 1303-10. [15]Guindon, Denis, "L'evolution du chomage structurel au Quebec: Un nouveau coup d'oeil." Mimeo. Quebec: Ministere des Finances, 1986. [16]Jewitt, Ian, "Justifying the First-Approach to Principal-Agent Problems." Econometrica, September 1988, 1177-90. [17]Krueger, Alan B., "Incentive Effects of Workers' Compensation Insurance." Journal of Public Economics, February 1990, 73-99. [18]--, "Moral Hazard in Workers' Compensation Insurance." Mimeo. Princeton: Woodrow Wilson School, Princeton University, May 1988. [19]Lanoie, Paul, "Occupational Safety and Health: A Problem of Double or Single Moral Hazard." Journal of Risk and Insurance, March 1991, 80-100. [20]--, "The Case of Risk Premia for Risky Jobs Revisited." Economics Letters, February 1990, 181-85. [21]--. "The Impact of Occupational Safety and Health Regulation on the Incidence of Workplace Accidents: Quebec 1982-1987." Discussion Paper no 4189, Montreal Center of Research and Development in Economics (C.R.D.E.), 1989a. [22]--. "Government Intervention in Occupational Safety and Health: A Theoretical and Empirical Analysis." Ph.D. dissertation, Kingston, Ont: Queen's University, 1989b. [23]Rea, Sam Jr., "Workmen's Compensation and Occupational Safety under Imperfect Information." American Economic Review, March 1981, 80-93. [24]Rogerson, William P., "The First-Order Approach to Principal-Agent Problems." Econometrica, November 1985, 1357-67. [25]Ruser, John W., "Workers' Compensation Insurance, Experience-Rating and Occupational Injuries." Rand journal of Economics, Winter 1985, 487-503. [26]Statistics Canada. Catalogue 31-203, 71-202s, 72-002, 91-921, 91-922, 94-751, 94-754, Ottawa: Ministry of Supply and Services. [27]--. "Fixed Capital Flows and Stocks; Quebec 1960-87." Ottawa: Ministry of Supply and Services, 1988. [28]Viscusi, W. Kip, "The Impact of Occupational Safety and Health Regulation." Bell Journal of Economics, Spring 1979, 117-40. [29]--, "The Impact of Occupational Safety and Health Regulation, 1973-1983." Rand Journal of Economics, Winter 1986, 567-80. [30]-- and Michael J. Moore. Compensation Mechanisms for Job Risks, Wages. Worker's Compensation and Product Liability. Princeton: Princeton University Press, 1990.