2. 352 CONTEMPORARY ECONOMIC POLICY
demand to patronize businesses in the hospi- staff. The match between new workers and
tality industry (Corsun, Young, and Enz, 1996; their employers develop in the smoke-free
Glantz and Smith, 1997; Hyland, Cummings, environment, so that the smoke-free law does
and Nauenberg, 1999; Pakko, 2005). Other not represent any shock to the match. The
recent literature has measured private market employee separation rate in the long run could
provision of smoke-free environments to be higher, lower, or no different for restau-
accommodate consumer preferences and the rants in municipalities with smoke-free laws.
differential effect of smoke-free laws on res- In this study we use a panel data set with
taurant and bar profitability (Dunham and treatment and control groups to examine
Mariow, 2000, 2003, 2004). The purpose of the influence of local smoke-free laws on
this study was to examine how laws influence employee separations. A logistic regression
employee turnover, which is a key determi- of employee separation was estimated using
nant of operating cost for the industry. We data on employees of a franchiser of a national
examine whether the likelihood of employee restaurant chain operating in the state of
separation from a job at a full-service restau- Arizona over a 5-yr period. The chain operates
rant is influenced by the introduction or full-service restaurants serving alcohol, with
presence of a local smoke-free law, after con- seating for an average of 190 customers,
trolling for other factors that influence and offering mid-price meals. Dunham and
employee separation. Mariow (2000, 2003) note that the introduc-
tion of smoke-free laws has a varying impact
II. METHODOLOGY
on different segments of the restaurant indus-
try. Profitability is most impacted in restau-
The likelihood of a worker separating from rants with more seating, a larger share of
their job falls with tenure as workers learn sales from alcohol, and a larger share of seat-
more about the rewards and conditions of ing in the smoking-allowed section but is not
a particular job and employers learn more impacted by whether a restaurant is part of
about the performance of workers (Bartel a chain or independent.^ The restaurants we
and Borjas, 1977; Jovanovic, 1979; Viscusi, examine, with large seating capacity and alco-
1980). Personal characteristics such as educa- hol sales, have the characteristics of restau-
tion, age, health, and sex further influence the rants likely to be impacted by smoke-free laws.
likelihood of separation (Bartel and Borjas, The panel data set included payroll records
1977; Meitzen, Í986; Mincer and Jovanovic,
available for 2-wk pay periods for employees
1981; Royalty, 1998).
of 23 Arizona restaurants from April 1999 to
The introduction of a smoke-free law also April 2004 (see Table 1), as well as employee
could influence the match between an existing characteristics such as age, race, gender, and
worker and their job. The law may represent occupation. Each 2-wk employee pay period
a shock to the "match" for existing workers, served as a single observation. The restaurant
leading to an increase in separation rates.
franchiser allowed smoking at its restaurants
While many workers may prefer to work in
a smoke-free workplace, other job attributes in the absence of a municipal smoke-free law.
such as earnings from tips also may change Ofthe 23 restaurants, 12 were located in munic-
as a municipal smoke-free law is implemented. ipalities with a smoke-free law as of 2004. Pres-
Dunham and Mariow (2003) note that restau- ence of a smoke-free ordinance was obtained
rants negatively impacted by smoke-free laws from the Americans for Nonsmokers' Rights
are more likely to increase job responsibilities database (www.no-smoke.org) and confirmed
for their workers. Some existing workers may with the company management.
flnd the new bundle of job attributes inferior Three of the restaurants opened smoke free
to the previous arrangement. This is particu- (one in Tucson, one in Mesa, and one in Gil-
larly true of any group workers, such as work- bert). The smoke-free ordinance in Mesa also
ers who smoke, who may have found was implemented before April 1999, so that
a smoking-allowed work environment to be our database for the Mesa restaurants only
an amenity. contained observations for workers in the
The long-run effects of smoke-free laws on period after the smoke-free law was in effect.
employee separation rates are less clear, how- 2. Dunham and Mariow (2004) report that chain res-
ever. Over the long run, there is turnover in taurants offered more space for nonsmoking seating.
3. THOMPSON ET AL.: SMOKE-FREE LAWS AND EMPLOYEE TURNOVER 353
TABLE 1
Statistics for Arizona Restaurants
Date When Community
Location Opened County went Smoke Free
Restaurants in communities with smoke-free laws as of 2004
Mesa, Arizona (1) December 1992 Maricopa July 1996
Mesa, Arizona (2) November 1992 Maricopa July 1996
Mesa, Arizona (3) June 1993 Maricopa July 1996
Mesa, Arizona (4) November 1998 Maricopa July 1996
Tempe, Arizona (1) June 1994 Maricopa May 2000
Tempe, Arizona (2) April 1997 Maricopa May 2000
Chandler, Arizona November 1997 Maricopa October 2003
Gilbert, Arizona May 2002 Maricopa May 2001
Tucson, Arizona (1) September 1991 Pima October 1999
Tucson, Arizona (2) May 1994 Pima October 1999
Tucson, Arizona (3) March 1997 Pima October 1999
Tucson, Arizona (4) January 2000 Pima October 1999
Restaurants in communities without smoke-free laws as of 2004
Phoenix, Arizona (1) December 1992 Maricopa No
Phoenix, Arizona (2) May 1995 Maricopa No
Phoenix, Arizona (3) October 1995 Maricopa No
Pheonix, Arizona (4) June 2002 Maricopa No
Peoria, Arizona September 1993 Maricopa No
Scottsdale, Arizona December 1994 Maricopa No
Prescott, Arizona February 1996 Yavapai No
Glendale, Arizona August 1996 Maricopa No
Goodyear, Arizona October 2000 Maricopa No
Surprise, Arizona June 2001 Maricopa No
Sierra Vista, Arizona September 2003 Cochise No
Six remaining restaurants were in municipali- only after the restaurant's municipality imple-
ties that were not smoke free in April 1999, but mented a smoke-free law.^ The control group
then implemented a smoke-free law later in the analysis consisted of restaurant payroll
either in October 1999 (Tucson), May 2000 records during any period when the restaurant
(Tempe), or in October 2003 (Chandler). did not face a local smoke-free law, either
Given the relatively short tenure of restaurant because the municipality where the restaurant
workers (see Table 2), the 7 mo of preban data was located never had a smoke-free law or
for workers in Tucson restaurants and 12 mo because the law was not yet in effect. There
in Tempe are sufficient for preban and post- were 90,810 payroll records in the control
ban comparisons of separation rates within group.
restaurants. Age, gender, ethnicity, job tenure, occupa-
The two treatment groups used in the anal- tion, and separation date were obtained from
ysis included restaurant payroll records during company payroll system records. The payroll
any period when a restaurant operated under database did not include data on other per-
a local smoke-free law. Treatment Group I sonal characteristics of workers that could
included 14,927 postban payroll records from influence employee separation rates, such as
employees who worked at a restaurant both education level and marital status, or other
before and after the municipality where the factors that could influence worker reaction
restaurant was located implemented a smoke-
free law. For these workers, the introduction
of a smoke-free law represented a potential 3. Therefore, Treatment Group II included payroll
"shock" to their work situation. Treatment records for employees ofthe three restaurants that opened
smoke free, and employees of the nine restaurants in
Group II included 69,966 payroll records Treatment Group I who started working there only after
for employees who worked at a restaurant the restaurant became smoke free.
4. 354 CONTEMPORARY ECONOMIC POLICY
TABLE 2 job tenure squared, and personal characteris-
Summary Statistics tics (age, gender, and race/ethnicity), as well as
a variable indicating the presence of a smoke-
Standard free law. There also was a dummy variable for
Variable Mean Deviation
each restaurant to control for idiosyncratic
Probability of separation and tenure working conditions, and a dummy variable
% separating during 4.2 20.0 for each month-year from April 1999 through
the pay period April 2004 to account for season and business
Tenure (d) 539 632 cycle impacts. Some employees had two
Tenure squared (d) 685.343 1,603,303 employment spells at a restaurant, and each
Personal characteristics
spell was treated as separate members of the
Gender (%)
panel. A dummy variable was used to indicate
Male 47.8 50.0
the second employment spell. In the logistic
Female 52.2 50.0
regression, standard errors were adjusted for
Age (yr) 26.1 7.0
clustering on employee-specific identification
Race (%)
White 71.4 45.2
numbers.
Black 3.0 17.1 The second model pooled Treatment
Hispanic 20.3 40.2 Group I with the control group. The third
American Indian/Alaska 1.2 11.0 model pooled Treatment Group II with the
Native control group. For all three models, we pres-
Asian/Pacific Islander 0.1 2.9 ent coefficient estimates from the logistic
Not specified 4.0 19.5 regression as well as estimates of each varia-
Occupation (%) ble's marginal effect.
Server 54.8 49.8
Models 1 through 3 contain a single dummy
Hostess 17.1 37.6
variable indicating that an employee works at
Bartender 2.0 13.9
a restaurant in a municipality covered by
Kitchen 24.3 42.9
a smoke-free law during a particular pay
All other occupations 1.8 13.4
period. Coefficient estimates for the dummy
variable indicate that the average effect of
a smoke-free law on employee turnover in
to a municipal smoke-free ordinance, such as the years after the law is in effect. The models,
smoking behavior. Observations were avail- however, do not capture how the effect of
able for each 2-wk pay period for the entire smoke-free laws may vary over time. In partic-
employment period. Separation was assumed ular, such a law may have a differential effect
to occur at the date of each worker's last entry in the first few months it is in effect relative to
in the payroll record. Of the approximately the longer term. It is in this initial period when
9,300 workers in the payroll database, roughly most existing workers are facing a shock to
one-third were still employed with the fran- working condition in regards to secondhand
chiser at the end of the data set. smoke in the workplace. In the longer run,
The first model pooled observations from as there is a natural turnover in restaurant
members of Treatment Group I, Treatment staff, most workers will have joined the staff
Group II, and the control group. This model after the municipal smoke-free law was in
examined the impact of a smoke-free law on place. The long-run effect, if any, could differ
the probability of separation for all restaurant from the initial effect.
employees after a smoke-free law was in effect, We tested for this possibility by developing
regardless of when the workers began working an additional model. In this fourth model, we
at the restaurants. A variable indicating use the full sample from the first model (both
whether each employee's place of work oper- the treatment groups and the control group)
ated under a smoke-free law in a particular and replace the single dummy variable indicat-
pay period was assigned a value of 1 for all ing that the smoke-free law is in effect with
members of either T^reatment Group I or II a set of 13 dummy variables, which indicate
and a value of 0 for all members ofthe control the amount of time that had passed since
group. The probability of separation for the law went into effect. The first dummy in-
employees in any particular period was mod- dicates that the smoke-free law was in effect
eled as a function of an employee's job tenure. for one quarter or less; the second dummy
5. THOMPSON ET AL.: SMOKE-FREE LAWS AND EMPLOYEE TURNOVER 355
indicates that the law was in effect from 4 to In all three regressions, the probability of
6 mo (i.e., the second quarter after the law went separation fell with tenure in the job. At mean
into effect). There are 12 such dummy varia- values for tenure and tenure squared, the mar-
bles for the first 12 quarters the law is in effect, ginal effect of additional days of tenure
and afinaldummy variable indicating that the reduced the probability of separation. Fur-
law had been effect for more than 3 yr. ther, reestimates of the marginal effects at
higher levels of tenure (such as tenure =
III. RESULTS
2,000 d and tenure squared = 4,000,000 d)
indicated that the marginal effect of additional
Table 2 presents summary statistics for the days of tenure would remain negative. Thus,
workers in this sample. On average, 4.2% of the relationship between the greater tenure
workers separated from employment during and the probability of separation was negative
a single 2-wk pay period. The average tenure even for an average tenure of more than 5 yr
of workers at any time during the 5-yr period (2000 d is roughly 5.5 yr).
was 539 d, which is roughly 1.5 yr. More than The probability of separation also was
half of the employees were female. More than lower for workers in their second spell of
70% of workers were white, while roughly 20% employment at a restaurant in both the all
were Hispanic and 3% were African Ameri- workers and the new workers regression. This
can. The average age of workers was 26 yr could have occurred because workers in their
(standard deviation = 7 yr). More than half second spell were more familiar with the
of the workers were employed as servers, requirements of the job and managers also
about one-quarter as kitchen workers, one- were more familiar with the workers. No sta-
sixth as hosts, and a fraction as bartenders tistically significant difference was found in
or other occupations. the existing workers regression, but this may
Coefficient estimates from the logistic have simply refiected the smaller sample size
regression are presented in Table 3, along with available.
estimates on the marginal effect of each vari- The probability of separation was related to
able on the probability of separation. The ethnicity in all three regressions. Relative to
marginal effects are estimated at the mean white workers, the probability of separation
value for all variables. Coefficients for individ- was lower for Hispanic workers. Gender
ual month and restaurant dummies are not was not related to the probability of separa-
reported for brevity but are available from tion in any ofthe three regressions. In all three
the first author upon request. regressions, the probability of separation was
Results for all workers in Table 3 are for lower for other occupations than for the omit-
the case where Treatment Group I, Treatment ted category, kitchen workers. This makes
Group II, and the control group were pooled. sense because the other occupations category
The treatment group contains pay period includes managers who have longer tenure.
observations for all workers at a restaurant The probability of separation also was lower
operating under a smoke-free law, regardless for bartenders in two of the three regressions.
of" whether they joined the restaurant before Finally, in all three regressions, no statisti-
(Treatment Group I) or after (Treatment cally significant relationship was found
Group II) the smoke-free law went into effect. between the presence of a smoke-free law
Results for existing workers were for the and the probability of employee separation.
case where Treatment Group I and the control The coefficient on the "law in effect" variable
group were pooled. The treatment group con- is not statistically significant in any of the
tains pay period observations for workers at regressions. This implies that there is no effect,
a restaurant operating under a smoke-free on average, on the probability of separation in
law but who joined the restaurant staff before the years after a smoke-free law is adopted by
the law was implemented. Results for new a municipality. This finding, however, does
workers were for the case when Treatment not preclude an effect in the initial periods
Group II and the control group were pooled. after the smoke-free law is adopted when
The treatment group contains pay period the law provides an initial shock to the work-
observations for workers who joined the res- ing conditions of existing restaurant workers.
taurant staff only after the smoke-free law For example, there could be an initial increase
was in effect. in separation rates for existing workers after
6. 356 CONTEMPORARY ECONOMIC POLICY
00
vo (N ON
*
.
1
.0006
C»
o NO
o
i
.002
001
.011
003
lo
000
005
000
003
009
000
o 8
o o o o o o o o o o o o o
1 1
1
I r-
oo
ON r^ (N
o
(N
o m
(N S o
p p p
d o
8 8 8 p p
d d d d d d d d
(N — ON
o
p O
2 d d o
I I
o
O o
S -H
o p
vi p — ^r-
(N I— Ö Ö - ^ C J Ö Ö Ö Ö Ö Ö Ö O Ö
( N Ö Ö Ö C N Ö O O Ö I I I rnÖÖ
'7 II I I
H-S
_C3
s H 1 O NO
O p •* p -^ —
— - ^ O O O O O Í N o o o o o m o — o d d d — o -: oI I
^ d
o II I "^ — I I
—' m
I I
•o
o
8
í I c
:ca
ce
Ul
p
s s
o
< •T3
les
•T3
0) c c C
AU
8 o
ISO
w
ca u.
S3
.5 < ca a:
7. THOMPSON ET AL.: SMOKE-FREE LAWS AND EMPLOYEE TURNOVER 357
S the law is implemented, but several years later,
the long-run separation rate (for workers who
orl
«
joined the restaurant after the law was imple-
0»
o mented) may be lower in municipalities with
Ö
1
smoke-free laws. The effect of the smoke-free
law on separation rates varies through time,
but the average effect is zero.
f
To test this possibility, we estimated
a fourth regression, where the "law in effect"
rkei
1
u o * variable from the all workers regression was
*
e
(N
o
replaced with 13 dummy variables indicating
tin
'I s
Ö
the length of time that a municipal smoke-free
.a
X 1 law had been in effect. Joint significance tests
indicated that the coefficients on these 13
dummy variables were not jointly different
from zero. This suggests that there was no sig-
nificant effect on separation rates through
1 *
time, just as no average effect was identified
231
in Table 3.
o
o Coefficients for several individual dummy
1
variables were significant, however. In Figure 1,
we present the individual estimates from this
regression for these 13 dummy variables. In par-
ticular, we show the estimated marginal effect
orlkers
for each of the 12 quarterly dummy variables
and the 13th variable indicating that the
160,'
1.200
.041
O)
9.13
hJ .S
smoke-free law had been effect for more than
Z 1 1 3yr.
HO There is a statistically significant decline in
the separation rate for workers in first quarter
after the smoke-free law is implemented.'* In
ters
other words, workers are less likely to separate
B
,a
2 o * from their job in the first few months the law
u
!Ç was in effect. Point estimates remain negative
oei
o' m o q
O throughout thefirsteight quarters that the law
isti
U Ö 1
1 was in effect, and the negative marginal effect
is statistically significant in the sixth quarter.
i Point estimates alternate between negative
te
and positive values beginning with the ninth
s: a quarter and are not statistically significant.
These quarterly results do not show a consis-
Ail Work
.088*
.97
tent impact on separation rates.
oo
1 1 Over the longer run, we did not find evi-
dence of a relationship between municipal
1 smoke-free laws and separation rates. There
was no statistically significant relationship
between the introduction of municipal smoke-
free laws and the probability of separation
C beyond 18 mo.
O
cupat
4. We also examined whether the probability of sepa-
ration changed in the quarter before the local smoking ban
was implemented, as workers anticipated the coming
3 e o
change. We did not find a statistically significant change
ial
k-
tu
.G
'S in the chances of separation in the quarter before imple-
> Ô mentation.
8. 358 CONTEMPORARY ECONOMIC POLICY
FIGURE 1 aration rates across all quarters was not sig-
Marginal Effect of the Presence of a Local nificantly different from zero. Further, there
Smoke-Free Law on the Probability of was no evidence of a relationship between
Separation smoke-free laws and employee separation
beyond 18 mo.
0.004 Taken together, these results suggest that
municipal smoke-free laws did not change
o
£
0.002
0
• ' ' • /
/x /
• V
the separation rate for workers in the long
run. The laws also did not induce an increase
ñ -0.002
/ ^ / in employee turnover in the initial period after
•g) -0.004 implementation by disrupting the match
/
g -0.006 /
between existing full-service restaurant work-
-0.008 V ers and their employers. The latter result
implies that in the quarters after the imple-
mentation of a smoke-free law, the change
in bundle of working conditions—which could
Quarters Since Law In Effect
include changes in earnings from tips as well as
Note: ( • ) indicates statistical significance at the 10% the change in workplace smoking—did not
confidence level. increase the rate of separation among existing
workers overall.
IV. DISCUSSION AND CONCLUSIONS
By contrast, the limited evidence we did
find of a change in separation rates suggests
Previous economic research on smoke-free that restaurant workers are for a period more
laws has focused on how these laws affect likely to remain in their job after the imple-
demand for businesses in the hospitality indus- mentation of a smoke-free law, perhaps exper-
try or on the differential effect of smoke-free imenting with the new working conditions.
laws on restaurant and bar profitability. The These aggregate results do not imply that
current study is an effort to examine how municipal smoke-free laws have no impact
smoke-free laws influence the behavior of res- on the welfare of restaurant workers. The
taurant workers. In particular, we examined mix of working conditions after the introduc-
how adoption of municipal smoke-free laws tion of a smoke-free law may not match what
influenced employee turnover, a key determi- many workers would have chosen in the
nant of operating costs in the restaurant and absence of a regulation, even if the discrepan-
bar industry. We used a unique data set of em- cies did not appear to be large enough to drive
ployment records of a franchiser of a national up separation rates. Further, our analysis
restaurant chain operating 23 full-service res- of aggregate separation rates may mask an
taurants in the state of Arizona, a state where increase in separation rates for some groups
several municipalities have adopted smoke- of workers, such as smokers. But it is impor-
free laws. tant for business owners, who face the training
We found a statistically significant decline costs associated with employee turnover, that
in the probability of separation in the initial the implementation of municipal smoke-free
months after a smoke-free law was imple- laws did not lead to an increase in aggregate
mented as well as evidence that separation separation rates for restaurants of the fran-
rates were lower 16-18 mo after implementa- chiser we studied.
tion. However, there was no consistent pat- These restaurants, which provide mid-price
tern of either a decline or an increase in meals and serve alcohol, are common
separation rates after the implementation of throughout the United States. Several recent
a smoke-free law. No average effect was iden- studies have indicated that larger restaurants
tified in the years after implementation either serving alcohol are the types of restaurants
for "existing" workers who were employed at whose profitability may be more likely to be
the restaurant at the time of implementation affected by smoke-free laws (Dunham and
or for "new" workers who joined the restau- Mariow, 2000, 2003). Findings regarding
rant after implementation. While we found employee separation in these restaurants are
a statistically significant decline in separation therefore of general interest and do not merely
rates in two quarters, the joint effect on sep- represent a niche segment or lightly impacted
9. THOMPSON ET AL.: SMOKE-FREE LAWS AND EMPLOYEE TURNOVER 359
portion of the industry. This said, it is not Hyland, A. K., M. Cummings, and E. Nauenberg. "Anal-
known whether the same effect (or lack of ysis of Taxable Sales Receipts: Was New York City's
Smoke-Free Air Act Bad for Restaurant Business?"
effect) on separation rates would be found Journal of Public Health Management and Practice,
in other restaurants that offer a different 5, 1999, 14-21.
mix of services to a different customer base. Jovanovic, B. "Job Matching and the Theory of Turnover."
Future research on employee separation rates Journal of Political Economy, 87, 1979, 972-90.
needs to focus on workers in other segments of Law, M. R., and N. J. Wald. "Environmental Tobacco
Smoke and Ischémie Heart Disease." Progress in
the restaurant industry. Cardiovascular Diseases, 46(1), 2003, 31-38.
Future research on separation rates may be Meitzen, M. E. "Differences in Male and Female Job-
able to identify the effect of laws on specific Quitting Behavior." Journal of Labor Economics,
groups of workers, such as smokers. Such 4, 1986, 151-67.
research also may be able to gather data on Mincer, J., and B. Jovanovic. Labor Mobility and Wages.
Studies in Labor Markets. Cambridge, MA: National
additional factors that influence employee sep- Bureau of Economic Research, 1981.
aration, including employee education level Pakko, M. R. On The Economic Impact of Smoking Bans.
and family structure, or major life changes CRE8 Occasional Report No. 2005-2. Center for
faced by employees, such as graduation from Regional Economics, St. Louis, MO: Federal
Reserve Bank of St. Louis, 2005.
high school or college.
Royalty, A. "Job to Job and Job to Non-Employment
Turnover by Gender and Education Level." Journal
of Labor Economics, 16, 1998, 392-443.
REFERENCES Siegel, M. "Involuntary Smoking in the Restaurant Work-
place: A Review of Employee Exposure and Health
Bartel, A. P., and G. J. Borjas. Middle-Age Job Mobility: Effects." Journal of the American Medical Associa-
Its Determinants and Consequences, Men in the Pre- tion, 270, 1993, 490-3.
Retirement Years. Philadelphia: Temple University
U.S. Department of Health and Human Services. "The
Press, 1977.
Health Consequences of Involuntary Exposure to
Corsun, D. L., C. A. Young, and C. A. Enz. "Should NYC Tobacco Smoke." Atlanta, GA: Department of
Restaurateurs Lighten Up? Effects of the City's Health and Human Services, Public Health Service,
Smoke-Free Air Act." Cornell Hotel and Restaurant Centers for Disease Control and Prevention,
Administration Quarterly, 37(2), 1996, 8-9. National Center for Chronic Disease and Preven-
Dunham, J., and M. L. Mariow. "Smoking Laws and tion and Promotion, Office of Smoking and Health,
Their Differential Impacts on Restaurants, Bars, 2006.
and Taverns." Contemporary Economic Policy, Viscusi, W. K. "A Theory of Job Shopping: A Bayesian
18(3), 2000, 326-33. Perspective." Quarterly Journal of Economics, 94,
. "The Economic Incidence of Smoking Bans." 1980, 609-14.
Applied Economics, 35, 2003, 1935-42. Wakefield, M., L. Trotter, M. Cameron, A. Woodward,
. "The Private Market for Accommodation: G. lnglis, and D. Hill. "Association Between Expo-
Determinants of Smoking Policies in Restaurants sure to Workplace Seeondhand Smoke and
and Bars." Eastern Economic Journal, 30, 2004, Reported Respiratory and Sensory Symptoms:
377-91. Cross-Sectional Study." Journal of Occupational
Glantz, S. A., and L. R. A. Smith. "The Effect of Ordi- and Environmental Medicine, 45, 2003, 622-7.
nances Requiring Smoke-Free Restaurants on Res- Wells, A. J. "Lung Cancer from Passive Smoking at
taurant Sales: A Follow-Up." American Journal of Work." American Journal of Public Health, 88,
Public Health, 87, 1997, 1687-93. 1988, 1025-29.