The effect of selective INS monitoring strategies on the industrial employment choice and earnings of recent immigrants.

Davila, Alberto ; Pagan, Jose A.

I. INTRODUCTION

Recent work explores the labor-market impact of employer sanctions embodied in the Immigration Reform and Control Act of 1986.(1) An emerging consensus in this research suggests that the new law's effect on the number of undocumented workers in the United States has been minimal (e.g., Chiswick [1991]). Indeed, the Act's main effect may have simply been to redistribute the undocumented from industries closely monitored by immigration authorities to those weakly monitored. This view gains credence given the Act's provision making alien inspections on farms relatively expensive, and that the Immigration and Naturalization Service (INS) may concentrate its enforcement resources on industries with a high concentration of illegals per establishment (Hill and Pearce [1990]). Clearly, if the law aims to reduce the population of undocumented labor in the United States, then a de facto enforcement strategy that targets a subset of industries mitigates the impact and the effectiveness of immigration reform.

This study examines the consequences of the INS's industry-selective enforcement policy. We model the effect of this monitoring strategy using a two-industry economy where immigration authorities monitor only one sector. This theory suggests that the productivity-enhancing characteristics of workers belong in models measuring the employment and earnings effects of enforcement-induced wage penalties. With this in mind, we empirically estimate the effects of the new law on the distribution of labor in monitored and unmonitored industries employing a multinomial logit technique. Pooled earnings functions with data from workers before and after immigration reform are then estimated to determine the Act's wage impact in the monitored industry.

Our data set is the 1980 and 1990 Public Use Microdata Samples (PUMS). The groups used here include Mexican and non-Mexican recent male immigrants (those arriving within five years of the census date) as well as Mexican "early" male immigrants (those residing in this country more than five years before the census date) and a sample of non-Hispanic native white males. We distinguish among these groups because of the stylized fact that the recently arrived Mexican immigrant group contains the largest share of undocumented workers. As such, we hypothesize this group would be the most likely to be affected by immigration reform relative to the other groups.

We find that between 1980 (pre-Act) and 1990 (post-Act), the likelihood that recent immigrants of both groups joined the manufacturing industry (a relatively closely monitored sector) declined and the probability that such immigrants participated in the agricultural and service industries (weakly monitored sectors) increased. This trend was more pronounced for recent Mexican immigrants despite the fact that their manufacturing wages also declined following passage of the Act. Our findings suggest that the INS's selective enforcement strategy has had a differential impact on industry and immigrant labor markets, and that it appears to have shifted the group with the largest share of illegal workers to weakly monitored industries.

II. BACKGROUND

We hypothesize that the Act's provisions led the INS to adopt a selective enforcement policy across industries. Two economic incentives for this monitoring strategy come to mind. First, the Act increased the relative cost of INS farm inspections. According to the new law, immigration officials cannot inspect a farm without a search warrant or the owner's permission, a legal requirement not present for other industries. The law also gave illegal aliens working in American agriculture relatively lenient requirements to apply for U.S. citizenship. Arguably, this facet of the law increased the INS's relative monitoring cost of assessing the legal status of an immigrant farm worker.(2) The Act's concessions to the farm sector were included in the new immigration law to facilitate a ready supply of immigrant agricultural labor (see, for example, Zolberg [1990]).(3)

Second, the INS can optimally monitor only a subset of industries given a fixed enforcement budget, as previous studies suggest. Hill and Pearce [1990] define a monitoring model in which the INS targets industries with a high number of illegals per establishment. They note that manufacturing has the highest share of illegals per firm and the service and agricultural industries have the lowest.

INS apprehensions statistics may give us a glimpse at the INS cross-industry enforcement strategy adopted after the passage of the 1986 Act. From the preceding discussion, one would expect that the INS concentrated more of its efforts at apprehending undocumented workers in non-agricultural industries following the Act's passage. To investigate this possibility, we gathered monthly apprehensions data (contained in INS Form G-23.18), by industry, spanning the pre- and post-Act period. Figure 1 plots the ratio of on-site apprehensions in the agricultural sector vis-a-vis other industries between 1984 and 1989. Figure 1 clearly shows that there was a marked reduction in the relative number of apprehensions in the agricultural sector after the Act was passed in 1986. The "agriculture/other industries" apprehensions ratio fluctuated from 0.8 to 1.5 during January 1984 to October 1986. This ratio then fell to around 0.64 soon after the Act's passage, hovering around 0.20 during the 1987-1989 period.(4) This finding lends support to our a priori contention that the INS concentrated its enforcement efforts in non-agricultural industries.

Next, we sought to determine the relative apprehension risks of various ethnic/industry groups. First, the total number of apprehensions from January 1987 to December 1989 by industry (manufacturing, construction, agriculture, and services) and the two ethnic categories (Mexican and non-Mexican) were obtained using the INS apprehensions data mentioned above. These figures were then divided by the total number of employed immigrants in 1989, by respective ethnic group and industry, arriving in the United States between January 1987 and December 1989 using 1990 1-percent PUMS data. Finally, we used the services apprehension ratio for the non-Mexican immigrant group as the base category to estimate the relative apprehension risk index of the ethnic-industry groups.

Table I presents the risk indexes for Mexican and non-Mexican recent immigrants in four industries. There are some clear differences in the risk of apprehension among the industry-ethnic groups. For example, according to our estimates, a recent Mexican immigrant working in construction was 10.95 times more likely to be apprehended by the INS during the foregoing period than a non-Mexican worker employed in services.

Nevertheless, we note that the magnitude of these indexes may be potentially biased. An implicit assumption behind these figures is that the ratio of legal to undocumented immigrants is the same in each industry. In addition, Table I indicates that, for example, a recent Mexican immigrant working in manufacturing is fifty-eight times more likely to be apprehended than a non-Mexican immigrant in services. It could be argued that this relatively large apprehension risk for the Mexican group may be minuscule if the risk of apprehension of the non-Mexican group is insignificant. We note, however, that Hill and Pearce [1990] have found the apprehension risk to be non-negligible in many industries.(5)

TABLE I

Post-Act Estimated Relative Risk of Being Apprehended as an Illegal
Alien While Employed

Industry Mexicans Others

AGRICULTURE 7.85 1.19
CONSTRUCTION 10.95 1.19
MANUFACTURING 57.75 5.33
SERVICES 4.60 1.00

Source: INS Form G-23.18, Monthly Reports, various issues. See text
for an explanation on how the index is estimated.

Judging from these data we find support for the hypotheses that immigration authorities (1) increased their enforcement efforts in non-agricultural sectors following passage of the immigration reform act and (2) have also pursued a de facto selective enforcement policy across industries and immigrant groups since its passage. We next explore the theoretical implications stemming from these policies.

III. IMMIGRATION REFORM EFFECTS: A SIMPLE TWO-INDUSTRY MODEL

We conceptualize the impact of this differential enforcement strategy by expanding on a two-sector labor productivity model first developed by Hamermesh and Biddle [1993]. In particular, assume a perfectly competitive two-industry economy with two groups of workers: natives (N) and immigrants (I). Assume that workers have one characteristic, x, associated with higher productivity and that 50 percent of both natives and immigrants possess this trait. Furthermore, suppose that employers who hire undocumented workers can be penalized with a monetary fine, p. Say also that immigration authorities only monitor industry 2 and that employers in the monitored industry pass on p, the amount of the fine, to those workers they perceive are undocumented, i.e. recent immigrants.

The industry-level wages ([w.sub.j]) are determined by

(1) [w.sub.1] = [[Alpha].sub.1]x and

(2) [w.sub.2] = [[Alpha].sub.2]x - p,

where [[Alpha].sub.j] captures the return to x in the jth industry (j = 1,2). Both x and p are assumed to take the values of zero or one. Thus, for recent immigrants [w.sub.1] - [w.sub.2] = ([[Alpha].sub.1] - [[Alpha].sub.2]) x + 1, and for natives [w.sub.1] - [w.sub.2] = ([[Alpha].sub.1] - [[Alpha].sub.2])x. Finally, assume that workers choose industry 1 if and only if [w.sub.1] [greater than] [w.sub.2].

This formulation yields three possible cases:

Case 1. If -1 [less than] ([[Alpha].sub.1] - [[Alpha].sub.2]) [less than] 1 then all recent immigrants work in industry 1 while non-x-endowed natives choose either industry 1 or 2; x-endowed natives choose industry I if ([[Alpha].sub.1] - [[Alpha].sub.2]) [greater than] 0 and industry 2 otherwise.

Case 2. If ([[Alpha].sub.2] - [[Alpha].sub.1]) [greater than] 1 then immigrants and natives endowed with x select industry 2. Immigrants without x choose employment in industry 1 while their native counterparts choose either industry 1 or 2. Note that the industry targeted for employer sanctions also demands individuals who possess x. The x-endowed immigrants accept the wage penalty to obtain the benefits accruing from their productivity endowment.

Case 3. If ([[Alpha].sub.1] - [[Alpha].sub.2]) [greater than] 1 then all immigrants as well as x-endowed natives select industry 1 while non-x-endowed natives are indifferent between 1 or 2.

This model can be generalized to an economy patterned by multiple industries and workers with more than one productivity-enhancing trait. The main results, however, remain intact: illegal immigrants will work in industries associated with a high undocumented-related wage penalty only when the remuneration from their productivity-enhancing characteristics exceeds the wage penalty. The model allows for the possibility that industries will contain both legal and illegal workers. Empirically, it highlights the importance of controlling for labor characteristics to properly assess the impact of immigration reform on the industrial selection and wages of workers. We turn to an empirical estimation of the major predictions of this model.

IV. INDUSTRIAL EMPLOYMENT METHODOLOGY

Following the structure of the conceptual framework, the analysis here assumes that recent immigrants select industries according to their socio-economic status, personal characteristics, and individual tastes and preferences. In turn, the profit-maximizing employer's hiring decision depends on the human capital characteristics of the immigrant. As such, we employ a multinomial logit specification because this method allows us to predict a post-Act industry employment level given an immigrant's vector of characteristics.(6)

The probability that a recently arrived immigrant i is employed in industry j depends on a vector of i's exogenous variables ([Z.sub.i]), namely,

(3) [P.sub.i](Industry = j) = f([Z.sub.i]), i = 1, ... T, j = 1, ... J.

This unknown probability function is usually specified as a logistic conditional probability function such as

(4) [P.sub.ij] = exp([Z[prime].sub.i][[Beta].sub.j]) / [summation of] exp ([Z[prime].sub.i][[Beta].sub.j]) where j = 1 to J,

where [[Beta].sub.j] = Kx1 vector of coefficients for industry j. The normalization [[Beta].sub.1] = 0 assures identification of the coefficients in these vectors. Maximum likelihood estimates of the parameters can be obtained with any nonlinear optimization technique given the global concavity of the log likelihood function in the [[Beta].sub.j]'s.

To determine how the new law altered the industrial distribution of recent immigrants, we estimate separate multinomial logit models of industrial attainment (eq. 4) for 1980 Mexican and non-Mexican recent immigrants. We then use the 1980 coefficients [Mathematical Expression Omitted] to generate an industry distribution for 1990 recent immigrants. The technique assumes that identical actual and predicted distributions would exist net of the Act.

The predicted probability that the ith recent immigrant from the 1990 sample would be employed in industry j [Mathematical Expression Omitted], given his individual characteristics [Mathematical Expression Omitted], and using this industry's "probability density function" for the 1980 sample, can be calculated by

(5) [Mathematical Expression Omitted].

By adding the jth industry's predicted probabilities across the 1990 immigrant sample, the "Act-neutral" industrial distribution can be estimated.

The magnitude of 1980 and 1990 industry differences, by group, can take the form of

(6) [Mathematical Expression Omitted],

where I is the total number of immigrant workers in the sample, [I.sub.j] represents the number of individuals who are employed in industry j in 1990, and [Mathematical Expression Omitted] defines the (model) predicted number of individuals in industry j if no structural differences between 1980 and 1990 existed. This dissimilarity index may be interpreted as the percentage of individuals who would have to switch industries in order to equalize the actual and predicted 1990 distributions.

V. INDUSTRIAL EMPLOYMENT FINDINGS

The 1980 and 1990 PUMS are utilized to estimate the industrial distribution model in this section and the earnings functions in section VI.(7) The PUMS particularly suits this study because it provides a relatively large sample of recent immigrants along with socio-economic data for individuals before and after passage of the Act. Our universe in this first part of the discussion consists of employed male immigrants who arrived to the United States within five years of the respective census dates.

We divide the sample into Mexican and non-Mexican immigrants because we hypothesize that the Act had a differential impact on these groups. Recall that INS apprehensions statistics show that during 1987-1989, Mexican undocumented workers comprised the majority of the "apprehended while employed" category (see INS Form G-23.18), so we expect this group to be more affected by the new law (see Table I).(8) Also, we divide the sample into agriculture, construction, manufacturing, and services industry categories for consistency with INS apprehensions data.

Multinomial logit models, as specified in section IV, are estimated for separate 1980 Mexican and non-Mexican recent immigrant male samples. Table II presents the variable definitions. The model includes controls for each individual's human capital stock and socio-economic characteristics.(9) Note that the estimated model predicts industrial employment selection well.(10) The coefficient estimates are used to calculate the predicted probabilities for each individual from the 1990 PUMS immigrant sample (eq. 5), and the sum of these simulate the 1990 industrial labor distribution that would have existed in the absence of immigration reform.

Table III reports the actual and predicted industrial distributions for the separate immigrant groups. By performing a [[Chi].sup.2] goodness-of-fit test, the null hypothesis that these distributions are identical was rejected.(11) Note that the model substantially overpredicts the percentage of Mexican (non-Mexican) workers in manufacturing by roughly 74 percent (54 percent). That is, ceteris paribus, the share of recent immigrants that chose manufacturing in 1990 relative to 1980 declined. Recall that manufacturing is the sector where both groups have the highest risk of apprehension while employed, and that Mexican immigrants are more likely to be apprehended in this sector than non-Mexicans (see Table I).

By the same token, both immigrant groups increased their employment shares in agriculture and services: the model underpredicts the number of immigrants in these sectors in 1990. For example, in 1990 about 35.42 percent of Mexican immigrants were employed in agriculture while our model predicts that share to be about 30.35 percent. Note also the consistency with our hypothesis that new penalties under the Act motivated some immigrants to select weakly monitored industries (see Table II).(12)

While these results are supportive of our expectations, they are also consistent with alternative hypotheses. That is, employment shifts occurred for many population groups during the 1980s for reasons independent of immigration reform (see, for example, Costrell [1994]). To strengthen our results, we estimated eq. (4) using a sample of immigrants who have been in the United States for more than five years preceding the 1980 and 1990 census dates (henceforth "early immigrants"). Our aim was to control for possible macroeconomic shocks which may have driven our previous results by using a base group that might have been equally affected by these shocks. We thus [TABULAR DATA FOR TABLE III OMITTED] predict the industrial distribution of recent immigrants, using the coefficients estimated from a multinomial logit model of early immigrants, for the Mexican and non-Mexican groups separately. We perform this procedure separately using the 1980 and 1990 PUMS's samples.

Our findings, available upon request, are particularly striking for Mexicans in manufacturing. Our model predicts roughly similar 1980 actual and predicted distributions between the recent and early Mexican groups. In 1990, however, the model overpredicts the percentage of recent-immigrant Mexicans employed in manufacturing (the estimated 1990 predicted employment share is 16.70 percent as compared to the actual 1990 share of 12.5 percent). The results support the hypothesis that recent Mexican immigrants avoided employment in the more closely monitored industry relative to early Mexican immigrants.

It could be argued, however, that employers may perceive any Mexican employee as a potential undocumented worker. To address this concern, we drew samples of native-born non-Hispanic whites (assuming these workers, from an employer's view, have the lowest probability of being undocumented) and recent Mexican immigrants from both the 1980 and 1990 PUMS. We then pooled these samples by the respective PUMS. Next multinomial logits for 1980 and 1990 containing a dummy variable identifying recent Mexican immigrants and the same control variables used in the previous logit regression were estimated. Admittedly, this empirical method departs from the one we use above; nevertheless, we opt to use this alternative methodology because it gives us the opportunity of reporting marginal effects for each group net of confounding influences, and it allows us to test how well our results hold up to alternative formulations. These results are presented in Tables IV and V.

The reported coefficients of the multinomial logit can be interpreted as the effects of the independent variables on the likelihood of choosing a particular industry over a base industry, in our case, agriculture. The coefficients of the variable MEXICAN in Table IV show that in 1980 recent Mexican immigrants were more likely to be found in manufacturing and construction when compared to non-Hispanic [TABULAR DATA FOR TABLE IV OMITTED] white natives. Expressing this in terms of marginal effects (Table IV in bold), we find that Mexicans were 5.8 percent more likely to be employed in construction and manufacturing than non-Hispanic natives of comparable characteristics during the period before the Act, and less likely to choose employment in agriculture.(13)

Table V reports the estimated coefficients of the logit model for the 1990 PUMS sample. The most important result is that, ceteris paribus, Mexican immigrants were more likely to be found in agriculture than manufacturing, when compared to non-Hispanic whites. This result is statistically significant and consistent with our assertion that recent Mexican [TABULAR DATA FOR TABLE V OMITTED] immigrants avoided employment in the manufacturing industry after the passage of the Act (when compared to the control group). Also, the marginal effect of the variable MEXICAN on the probability of choosing agriculture and manufacturing gives further support to this view; it is positive for agriculture and negative for manufacturing.

Finally, to capture the degree by which the two groups changed their industrial employment choice between 1980 and 1990, we estimate an industrial distribution dissimilarity index (eq. 6). These results are presented in the bottom of Table III. We find that 10.83 percent (9.71 percent) of Mexican (non-Mexican) immigrants would have to switch industries to match the actual and predicted 1990 industrial distributions. Thus, we support the proposition that employed recent Mexican immigrants were relatively more affected by the Act's passage in terms of industry mix, net of confounding effects.

The combined results from this section lead us to the following conclusion. After controlling for the factors that affect industry choice, as suggested by our theoretical model, we find that Mexican and non-Mexican recent immigrants appear to have selected relatively more employment in the weakly monitored (services and agriculture) than the heavily monitored industries (manufacturing and construction) in the post-Act era.

VI. EARNINGS FUNCTION RESULTS

We have found support for the hypothesis that the Immigration Reform and Control Act had an impact on the industrial distribution of recent immigrants. The conceptualization of the impact of the Act on undocumented labor markets suggests that the earnings dimension should also be considered. We thus turn to an empirical investigation of an Act-induced wage penalty.(14)

We employ the familiar human capital earnings function to conduct this part of our empirical analysis. Table II contains the definitions of the variables used in the earnings functions. Before discussing our empirical results, two points should be made about our data. First, variables in the multivariate analysis are adjusted for consistency purposes across the two PUMS.(15) Second, the dependent variable, annual earnings, is transformed into a real annual earnings measure using Consumer Price Index (1982-1984) data.

As noted in section V, influences other than immigration reform could serve as alternative explanations for empirical findings which rely on tracking the labor-market experience of recently arrived Mexican immigrants alone. For example, Costrell [1994] notes that "employment shifts in the 1980s from high-wage industries to low-wage industries had a large shift-share effect on average pay growth." In related evidence, Murphy and Welch [1991] show that product demand shifts (associated with large trade deficits) during the 1980s led to a sharp decline in manufacturing employment and a shifting of employment towards other sectors in the economy. Of course, these macroeconomic effects may be responsible for an observed decline in the earnings of recent Mexican immigrants in the manufacturing sector. It follows that we should identify a relative decline in the earnings of this group by selecting a comparable base group who a priori should have had a similar labor market experience.

Therefore, we estimated a pooled earnings function that includes recent Mexican immigrants and non-Hispanic white males from the 1980 and 1990 PUMS (a sample similar to that employed in section V).(16) This is done in an attempt to "control" for non-Act movements in industry wages between 1980 and 1990. The regression includes a set of control variables, as well as year, ethnic group, and industry interaction terms.

Table VI presents the estimated equation results. Note that all the human capital and socio-economic variables' coefficients have the expected signs. In particular, the coefficient for MEXICAN x 1990 x MANUFACTURING is negative and statistically significant. That is, Mexican immigrants experienced a relatively large earnings decline in manufacturing in 1990, when compared to native non-Hispanic whites employed in the same industry during that year. The same result is evident in the construction sector (see the MEXICAN x 1990 x CONSTRUCTION coefficient), the other sector where recent Mexican immigrants face a high risk of [TABULAR DATA FOR TABLE VI OMITTED] apprehension. Our findings also suggest that, when compared to the base group, recent Mexican immigrants achieved relative wage gains in those industries associated with a low apprehensions-while-employed risk (agriculture) or at least their earnings did not decline as much (e.g. services) as in those industries where apprehension risk is high.

In sum, our earnings functions estimates support our expectations. That is, we provide evidence that the earnings of recent male Mexican immigrants fell on a relative basis compared to those of native male non-Hispanic whites, particularly in those industries that were highly monitored by the INS during the 1980s.

VII. CONCLUDING REMARKS

We advance the hypothesis that the INS's policy of targeting industries where the costs per apprehension are relatively low may have the effect of turning weakly monitored industries into "safe-havens" for undocumented recent immigrants. Using the 1980 and 1990 PUMS, we show that, ceteris paribus, immigrants seem to have moved from construction and manufacturing (strongly monitored sectors) into agriculture and services (weakly monitored sectors). Our earnings function findings are also supportive of a relative earnings penalty against recent Mexican immigrants working in relatively monitored industries. This study thus adds to the research that has questioned the effectiveness of the Immigration Reform and Control Act. While our results say little about the total employment impact of the INS's selective enforcement strategies, our findings do suggest that such policies may allow some undocumented workers to be (and stay) employed in industries where they have a low probability of being detected.(17) On the other hand, our results could also suggest that the Act has been partially successful in reducing employment opportunities for recent immigrants most likely to be undocumented in the relatively high-paying manufacturing sector. The implications stemming from this justification for adopting a selective monitoring strategy are worthy of future investigation.

TABLE II

Definition of Variables

EARNINGS = total real annual earnings. AGE = age of individual. EDUCATION = number of years of education. HOURS = total annual number of hours worked. MARRIED = 1 if married; 0 otherwise. DISABLED = 1 if disabled; 0 otherwise. NORTHEAST = 1 if region of residence is northeastern U.S.; 0 otherwise. SOUTH = 1 if region of residence is southern U.S.; 0 otherwise. NORTH CENTRAL = 1 if region of residence is north central U.S.; 0 otherwise. WEST = 1 if region of residence is western U.S.; 0 otherwise. AGRICULTURE = 1 if employed in agriculture; 0 otherwise. CONSTRUCTION = 1 if employed in construction; 0 otherwise. MANUFACTURING = 1 if employed in manufacturing; 0 otherwise. SERVICES = 1 if employed in services; 0 otherwise. MEXICAN = 1 if individual is a recent Mexican immigrant; 0 otherwise. 1990 = 1 if individual belongs to 1990 PUMS; 0 if individual belongs to 1980.

1. The Act was signed by President Reagan on 6 November 1986, although full enforcement of the new law did not start until 1 June 1988. The Act made it unlawful for firms to knowingly hire undocumented aliens, extending this ban to all employers regardless of the number of workers hired. Previous to the Act, employers were not fined for employing undocumented labor although it was illegal for employers to transport and/or harbor undocumenteds (see Zolberg [1990]).

It is also of interest to note that the employer penalty for hiring undocumenteds ranges from $250 to $2,000 per alien, but repeat offenders could be fined up to $10,000 per worker or receive a criminal penalty of $3,000 with six months imprisonment. The law also afforded the INS more resources to improve its enforcement powers. For an overview of this literature, see the U.S. Department of Labor's [1991a; 1991b] first and second reports on the implementation and impact of employer sanctions.

2. Immigrant farm labor employed for at least ninety days from 1 May 1985 to 1 May 1986 were given the opportunity to become lawful temporary residents of the United States and, subsequently, permanent residents after five years.

3. Also, shortages in farm labor in the fiscal years 1990 through 1993 would allow for more immigrants, who after three years as temporary residents could become permanent residents.

4. The average monthly ratio in the pre-Act period was 1.062 (s.d. = 0.197; 6,984.4 and 6,621.5 average apprehensions per month in agriculture and all other industries, respectively) and 0.200 after November of 1986 (s.d. = 0.122; 470.7 and 2,380.1 average monthly apprehensions in agriculture and all other industries, respectively).

5. For this time period, apprehensions by industry and group were as follows. For the Mexican group: 24,357, 10,313, 45,877, and 14,750 were apprehended in agriculture, construction, manufacturing, and services, respectively. For the non-Mexican group: 1,841, 1,941, 11,071, and 14,384 were apprehended, respectively, in the aforementioned industries.

6. For a similar empirical application, see Brown, Moon, and Zoloth [1990].

7. The 1990 census contains a "weight" factor for sampled individuals. We estimated industrial distributions using this factor and found that the unweighted distribution was very similar for our samples.

8. It can be argued that the heterogeneity of the non-Mexican sample may bias our distribution results for this group. We replicated our results using non-Mexican Hispanic and "Other" immigrant dummies in our logit estimation but no significant changes in the predicted industrial distributions were observed. This was the lowest level of group disaggregation that we could attain given the way the INS collects apprehensions data.

9. The explanatory variables included a constant term, age, years of education, marital status, and controls for region of residence. These intermediate results are available upon request from the authors.

10. The [[Chi].sup.2] statistics (74.428 and 222.25 for the Mexican and non-Mexican sample, respectively), which test the hypothesis that all the coefficients are zero, were highly significant (less than 1 percent level). This implies that our model predicts industry much better than the naive sample proportions approach.

11. The [[Chi].sup.2] goodness-of-fit statistics were 230.51 and 543.17 for the Mexican and non-Mexican samples, respectively. They were estimated by

[Mathematical Expression Omitted],

where [I.sub.j] and [Mathematical Expression Omitted] represent the total actual and predicted number of immigrants employed in industry j, the null hypothesis being that the two distributions are identical. The null is rejected at the 0.005 level of significance for both distributions. The critical [Mathematical Expression Omitted].

12. The statistical significance of these results needs to be addressed. For this purpose, we estimated pooled (1980 and 1990) multinomial logit models with controls for the same exogenous factors as in the previous analysis, including a year dummy (1990 = 1; 0 otherwise). We then calculated the marginal probabilities at the variable means by using

[Delta][P.sub.j] / [Delta][Z.sub.i] = [P.sub.j]([[Beta].sub.j] - [summation over j] [P.sub.j][[Beta].sub.j]).

The 1990 effects are statistically significant across both groups of immigrants in every industry. These results are available upon request from the authors.

13. The partials were estimated by

[Delta][P.sub.j] / [Delta][Z.sub.i] = [P.sub.j]([[Beta].sub.j] - [summation over j] [P.sub.j][[Beta].sub.j]).

These were evaluated at the sample mean of each variable. Note that the marginal effects (derivatives) reported in Tables IV and V do not necessarily have the same sign as the variable coefficients. The [Beta]'s capture the log-odd ratios of being employed in a particular industry as opposed to agriculture while the derivatives capture the marginal effects of the regressors on the probability of being employed in the jth industry (see Greene [1993, 666-67]).

14. Employer fines become a direct cost of hiring illegal workers, and enforcement provisions increase the probability that an employer breaking the law will be detected and fined by the INS. The risk of paying a fine would be expected to be passed on to illegals in the form of lower wages. Also, some researchers contend that firm-specific investment losses from worker deportations existed before the law, and that monopsonistic practices may have also reduced the earnings of undocumenteds (see Davila, Bohara, and Saenz [1993] and Pagan and Davila [1996]).

15. The 1980 education variable had to be recoded for consistency purposes given that the 1990 census defines schooling categories differently.

16. Ideally, as suggested by one referee, one would like to compare the labor market experience of recent Mexican immigrants with that of recently employed non-immigrants. This would control for the "last-hired-first-fired" effect that may have occurred with the contraction in manufacturing employment during the 1980s. However, data limitations do not allow us to test this interesting hypothesis.

17. In this regard, it is of interest to note recent INS policies and their potential economic effects. The INS has recently adopted a plan that places relative emphasis on enforcement in the Southwest border region, particularly in El Paso, Texas and San Diego, California, with increases in border patrol personnel and upgrades in border-patrol technology and equipment (see U S Department of Justice [1994]). The plan's success hinges on the implicit assumption that the undocumented Mexican worker will not seek alternative ports of entry to those (El Paso and San Diego) which have traditionally been popular. As we have shown, however, employers and recent immigrants showed a high degree of flexibility during the early years of the post-Act period, putting the INS's current plans to safeguard the United States border into further question.

REFERENCES

Brown, Randall S., Marilyn Moon, and Barbara S. Zoloth. "Occupational Attainment and Segregation by Sex." Industrial and Labor Relations Review, 33(4), 1990, 506-17.

Chiswick, Barry R. "Illegal Immigration and Immigration Control," in U.S. Immigration Policy Reform in the 1980's: A Preliminary Assessment, edited by Francisco L. Rivera-Batiz, Selig L. Sechzer, and Ira N. Gang. New York: Praeger, 1991, 45-64.

Costrell, Robert M. "Accounting for the Causes and Consequences of Industrial Employment Shift." Industrial Relations, July 1994, 346-64.

Davila, Alberto. "Immigration Reform and the Earnings of Mexican, Hispanic, and Other Immigrants." Working paper, University of New Mexico, 1994.

Davila, Alberto, Alok K. Bohara, and Rogelio Saenz. "Accent Penalties and the Earnings of Mexican Americans." Social Science Quarterly, December 1993, 902-16.

Greene, William H. Econometric Analysis New York: Macmillan, 1993.

Hamermesh, Daniel, and Jeff E. Biddle. "Beauty and the Labor Market." Working paper, University of Texas-Austin and Michigan State University, 1993.

Hill, John K., and James E. Pearce. "The Incidence of Sanctions Against Employers of Illegal Aliens." Journal of Political Economy, 98(1), 1990, 28-44.

Murphy, Kevin M., and Finis Welch. "The Role of International Trade in Wage Differentials," in Workers and their Wages, edited by M. Kosters. Washington, D.C.: AEI Press, 1991, 39-69.

Pagan, Jose A. and Alberto Davila. "On-the-Job Training, Immigration Reform, and the True Wages of Native Male Workers." Industrial Relations, 35(1), 1996, 45-58.

U.S. Department of Justice. Monthly Report of Deport-able Aliens Found in U.S. by Nationality, Status at Entry, Place of Entry, Status when Found. INS Form G-23.18, various issues.

U.S. Department of Justice. Attorney General and INS Commissioner Announce a Two-Year Strategy to Curb Illegal Immigration. Washington, D.C.: Immigration and Naturalization Service Press Office, February 1994.

U.S. Department of Labor. Employer Sanctions and U.S. Labor Markets: First Report. Washington, D.C.: Bureau of International Labor Affairs, 1991a.

-----. Employer Sanctions and U.S. Labor Markets: Second Report. Washington, D.C.: Bureau of International Labor Affairs, 1991b.

Zolberg, Aristide R. "Reforming the Back Door: The Immigration Reform and Control Act of 1986 in Historical Perspective," in Immigration Reconsidered: History, Sociology, and Politics, edited by Virginia Yuns-McLaughlin. Oxford: Oxford University Press, 1990, 315-37.

ALBERTO DAVILA and JOSE A. PAGAN, Professor, and Assistant Professor, University of Texas - Pan American. Many thanks to Eleanor Brown and to two anonymous reviewers for their insightful comments. We thank Jack Diego for his research assistance and the Center for Regional Studies for its support.