Printer Friendly

Is employment discrimination based on tobacco use efficient?


Tobacco use is costly to individuals who smoke or ingest its contents by chewing. It is costly in terms of the additional medical care assumed by its current and former users, in terms of lost work days, and in terms of reduced life expectancy. It has health consequences that are similar to several other behaviors such as over eating. lack of exercise, excessive alcohol consumption, or the consumption of a range of drugs. Furthermore, where there exists a public health system, as in Canada and much of Europe, or an insurance-driven private system, as in most large organizations in the United States, the result of these individual choices and behaviors is that some of the costs and consequences are passed on to others: this is a classic problem of externalities.

Our paper has two objectives. One is to examine the degree to which smokers generate excess medical costs relative to individuals with other morbidities. The second objective is to better understand the measurement of the excess medical costs associated with being a former smoker, in particular the time lags associated with improved health post-quit. The term "excess medical costs" is used to denote the costs associated with one or more of these conditions or behaviors relative to the costs of not having such conditions and being a "never smoker."

The cost issue has considerable importance in the public policy domain at the present time. Many employers in the United States--health institutions in particular--refuse to employ smokers. These employers are not content to ban smoking in the place of employment (Siegel 2011: Sulzberger 2011; Zamora 2012). More than half of all states prohibit hiring discrimination on grounds such as being a smoker, but 21 do not (McDaniel and Malone 2012). Hospitals in Florida, Georgia, Massachusetts, Missouri, Ohio, Pennsylvania. Tennessee. and Texas discriminate in their hiring decisions. (1)

This type of discrimination appears limited at the present time--primarily to hospitals. However, a number .of nonhealth-based institutions also discriminate, and such discrimination is seen as a potential means of reducing smoking (McDaniel and Malone 2012). But given the legality of the practice it is appropriate to examine its efficacy as a means of cost reduction for any type of employer. Indeed the Floridian city of Delray Beach recently approved a ban on hiring smokers (RT 2012).

The hiring policy is motivated by considerations that go beyond the potential health damage associated with second-hand smoke (SHS). The dominant belief in the medical community at the present time is that even small reductions in tobacco-related toxins in ambient air can reduce the potential for a range of cardio-pulmonary morbidities--particularly myocardial infarction (Pell et al. 2008). This is also the definitive conclusion of the Institute of Medicine (IOM 2009). (2) However, SHS damage can be controlled by prohibiting smoking in the environs of the employment location; it is not necessary to discriminate against smokers at the hiring point.

Accordingly, employment discrimination has two objectives: one is to protect the individual from herself--smoking is bad for health and smokers do not behave in their own best interests. Second, smoking is costly to the employer: because smokers incur more sick days than nonsmokers they have lower productivity, and also because of higher insurance costs incurred by the employer wherever an insurance plan is on offer to employees. The first motive could be termed paternalism, the second motive employment efficiency or cost minimization. A good example of the stated reasoning behind the first type of discrimination is to be found in Cleveland Clinic (2007). (3)

The paternalist motive for discrimination runs counter to the rational addiction model of Becker and Murphy (1988). They perceive smokers as individuals who may be willing to trade off nearer-term pleasure with longer-term poorer health. Nonetheless, a number of frameworks have been developed relatively recently that seek to give an intellectual underpinning to intervention in individual decisions, even in the absence of externalities. (4)

In contrast to this paternalist approach, employers in those states where discrimination is legal have motives other than a job applicant's well-being--productivity enhancement, or cost reductions through fewer lost days of work. In large organizations, the additional risk and insurance costs associated with employing poorer-health individuals are offset by the reduced costs associated with employing healthier individuals. But improved average health can still be achieved by discriminating.

Thus, whether discrimination against smokers is practiced for reasons of paternalism or cost minimization, it is vital to understand the likely "value" of such discrimination in the context of an environment where potential employees may be current smokers, former smokers, or suffer from one or a number of other morbidities. Our objective is to contribute to understanding this picture by examining one large component of such costs--hospitalizations--and to improve our understanding of the costs associated with individuals who have already quit smoking.

As an alternative to hiring discrimination, employers could choose to implement a compensatory policy of wage discrimination that favors those with better health. Cawley (2004) finds that obese individuals earn less than thin individuals after correcting for observable productivity characteristics. Bhattacharya and Bundorf (2009) find that this relationship holds only for organizations that provide health insurance to their employees, and that such shifting is absent in organizations without insurance.

In contrast to discrimination in wages or at the point of hiring. U.S. employers have limited legal ability to vary health premiums on the basis of health conditions (Bhattacharya and Bundorf 2009, p. 148 and references therein). This limitation places increased importance on wage and hiring discrimination, and such discrimination has a relatively low implementation cost where smokers are concerned: screening costs are small, in part because the pool of applicants is reduced as a result of smokers self-selecting out of the application process. The antismoker policy can be enforced with body fluid tests. Cotinine is a metabolite of nicotine and is present in blood, saliva, and urine. It has a half-life of about one day in these media. Cotinine is also present in hair in measurable amounts, and is detectable for several weeks after nicotine has been ingested. Hair-based tests thus permit the detection of smoking patterns even if an individual has not smoked for several days and might thus pass a fluid-based test (see Florescu et al. 2009).

However, our results indicate that a hiring perspective focusing only upon avoiding smokers is incomplete, in that it disregards the health costs associated with being a former smoker, and in particular a recent quitter. We find that the medical costs incurred by recent quitters exceed, on average, the costs associated with being a current smoker, because considerable time is required before medical costs attributable to former smoking decline significantly. Consequently, given that substantial transition costs may be associated with quitting, it is important to understand and characterize them accurately. The IOM report on smoking (2009, p. 15) states that there is a long-term benefit to quitting, but that such benefit may be accompanied with near term costs:
  Smoking cessation has been associated with reduced risk of coronary
  heart disease. The speed and magnitude of risk reduction after
  smoking cessation, however, have been debated (references). Some
  studies found that risk could decline to that of a lifelong
  nonsmoker (references), and others have suggested that some
  residual excess risk remains (references). On
  the basis of a systematic review of 20 cohort studies. Critchley and
  Capewell (2003) estimated that there was a 36% reduction in
  Mortality in patients with coronary heart disease who quit smoking
  Compared with those who continued smoking.

Before proceeding to our analysis, it is worth noting that our paper is not directed at a determination of the equity consequences of discrimination. As lower-income/education individuals have smoking rates that are a multiple of the rates for higher income/education individuals we take it as a given that employment discrimination based upon tobacco use is inequitable in the traditional sense. An extensive treatment of the ethical aspects of this form of discrimination is given by Voigt (2012).

The paper proceeds as follows. In the next section we describe the data and explain why it is critical to examine the behavior of former smokers. Section III explores an appropriate econometric specification of the problem at hand. Section IV presents results and performs robustness tests. Section V concludes.


Data for our analysis are from the Canadian Community Health Surveys (CCHS) of 2003, 2005, and 2007. These cross-sectional surveys are nationally representative, are the most comprehensive survey on health in Canada, and contain in excess of 130,000 records each year for a very detailed questionnaire. In .addition to a full set of socio-demographic information, individuals are surveyed on their use of the hospital system, their smoking patterns past and present, whether they suffer from a chronic condition, their consumption of alcohol, their eating and exercise patterns, their weight and height, their mental health, and so forth.

The outcome variable in our analysis is the number of full day-night cycles spent by an individual in hospital. The CCHS asked two questions regarding respondents' hospitalization over the preceding 12 months: First, In the past 12 months, have you been a patient overnight in a hospital, nursing home or convalescent home?" Then, for those who said yes to the first question, they were asked: "For how many nights in the past 12 months?" We use the responses to these two questions to construct the number of hospital days.

We exclude outpatient and day visits as outcome variables, as these are often of a preventive nature and do not reflect poor health. For example, blood tests, mammography, and colonoscopies are frequently carried out in hospitals, and may reflect the fact that a patient has a full periodic annual medical check-up, not that they are suffering from a particular medical condition. Furthermore, day visits do not normally require absence from work for a full day. Work days lost are important for the employer, and these latter are more highly correlated with full day-night hospital visits than just day visits.

The value of considering all-cause hospital days as .a cost measure, as opposed to hospital days directly associated with the occurrence of particular conditions such as myocardial infarction, is that smoking (and other behaviors) may impact a wide range of morbidities. In addition to getting a broader understanding of the impacts, the use of hospital days regardless of the immediately stated cause captures whatever elements of competing risk may exist among these morbidities.

That said, hospital days are not an exact measure of financial costs incurred at the hospital, or by the insurer: a heart bypass operation or a kidney transplant is more costly than an individual being "under observation." (5) Unfortunately, such detailed information on the specific type of treatment received while in hospital is not available in the CCHS surveys, and so the dollar cost of hospital stays that are associated with smoking and other conditions is not in our data. Nonetheless, if our main perspective is that of an employer, such costs are secondary. It is days lost from work, regardless of their financial cost within the medical system, that concerns an employer. Days in hospital may be a good measure of such lost work days.

In addition to its size and depth, an advantage to using this Canadian Survey is that there exists a universal system of medical insurance that has very little variation from province to province, since each province's coverage is governed by the common federal legislation in the Canada Health Act. As a consequence, the data should manifest little differential moral hazard behavior that can characterize data where individuals may or may not be insured, such .as Occurs when private insurance dominates the health sector. Behavior changes following the availability or purchase of private insurance is discussed by Bhattacharya and Sood (2011). (6)

The data sample we use encompasses all individuals surveyed, regardless of their current or former smoking status. On smoking behavior, the survey asked respondents about their smoking status: current smokers (daily versus occasional), former smokers, and never smokers. Current smokers are asked about the number of years for which they have been smoking. Former smokers are asked about their age when they first smoked and the number of years since they stopped smoking. To obtain a variable that measures the number of years smoked by a former smoker, we simply subtract the age at which the individual first smoked plus his "number of years since stop smoking" from his current age.

The "nonsmoker" group in our data consists only of never-smokers. It is not assumed that the excess costs of being a former smoker decline to zero over any horizon, nor is it assumed that these post-quit costs evolve linearly. As we shall see, the costs of being a former smoker, are highly non-linear over time, and excess costs characterize this group for up to 6 years, or even more in some cases. Higher costs for quitters than current regular smokers could arise on account of withdrawal costs or because of selection--those whose health is worst may be over represented in the quit group.

Our data sample consists of respondents aged 25-65 as we want to focus upon those of working age. In the econometric analysis we will also present estimates for a wider age group in addition to the estimates for this prime age group. Table 1 presents the characteristics of this sample. Respondents are quite evenly distributed over four age groups. Given the age categories, it is not surprising that a large proportion of the sample respondents (80%) are married, and the average family size is 2.4 persons. There is slightly higher female representation in the sample than male (54% vs. 46%). Immigrants account for 20% of the sample. For this specific age group 25-65, most of the respondents possess some post-secondary education; 17% has completed secondary school and those who have less than secondary education account for 16% of the sample.
                                  TABLE 1

                             Summary Statistics

Variable                                 Mean   Std.     Number of
                                                Dev.   Observations

Proportion of being hospitalized         0.083  0.276     239,920

Number of days for a patient             4.426  4.783      18,894

Number of days for the whole sample      0.347  1.791     232,009

Current smoker                           0.274  0.446     232,714

Former smoker                            0.425  0.494     232,714

Nonsmoker                                0.297  0.457     232,714

Current daily smoker                     0.226  0.418     232,714

Current occasional smoker                0.049  0.215     232,714

Daily former smoker (quit 2 years)       0.056  0.231     232,714

Daily former smoker (quit 3-5 years)     0.032  0.176     232,714

Daily former smoker (quit over 6 years)  0.181  0.385     232,714

Occasional former smoker                 0.155  0.362     232,714

Chronic disease                          0.571  0.495     232,975

Obese                                    0.204  0.403     233,036

Asthma                                   0.081  0.273     232,864

Heart                                    0.036  0.187     232,683

Hypertension                             0.151  0.358     232,453

Diabetes                                 0.050  0.219     232,808

Back pain                                0.233  0.423     232,797

Migraine                                 0.123  0.328     232,832

Other chronic conditions                 0.108  0.310     233,036

Physically active                        0.499  0.500     229,895

Binge drinking                           0.307  0.461     233,036

Age 25-34                                0.232  0.422     233,036

Age 35-44                                0.256  0.437     233,036

Age 45-54                                0.255  0.436     233,036

Age 55-64                                0.257  0.437     233,036

Student                                  0.045  0.207     233,036

Male                                     0.465  0.499     233,036

Married                                  0.802  0.398     232,525

Hhsize                                   2.431  1.201     233,036

Secondary school graduation              0.171  0.377     229,776

Some post-secondary                      0.067  0.250     229,776

Post-secondary graduation                0.606  0.489     229,776

Immigrant                                0.203  0.828     228,891

In terms of health-related behavior, 30% of respondents are never-smokers, 27% are current smokers, and the remaining 43% are former smokers. Thirty-one percent reported having had a drinking binge during the preceding 12 months, while almost 50% stated that they were either moderately or highly physically active.

Regarding health outcomes, 8.3% of people aged 25-65 were hospitalized over the preceding 12 months, with each patient spending on average 4.4 days. (7) For the whole population aged 25-65, the number of overnights at hospital is 0.35 for each person. Figure 1 describes the frequency of stays. The frequencies for each of the three survey years are strongly similar and little is to be gained by viewing them separately. The dominant characteristic of this frequency function is the number of individuals who record no hospital stays during the year, and this characteristic is what determines an appropriate estimator (see below).


The incidence of what is described as a "chronic condition" is high. In these surveys it includes almost 20 different morbidities, (8) and 57% of the sample suffers from at least one. The incidences of a limited number of these conditions in our sample are given in Table 2. Back pain tops the list with 23% of the sample being afflicted. The second most prevalent condition is hypertension with a prevalence of 15%. This hypertension prevalence is three times that .of diabetes, five times that of heart diseases, and twice that of asthma. About 10% of the sample has some chronic conditions other than these. The prevalence of these conditions is also broken down by category of smoker.
                        TABLE 2

       Burden of Chronic Disease by Smoking Status

                  Current  Former  Non-smoker  Total
                  Smoker   Smoker

Having a chronic    0.602   0.594     0.508     0.57

Obese               0.176   0.226     0.197    0.204

Asthma              0.084    0.08     0.077     0.08

Heart               0.038   0.043     0.023    0.036

Hyper               0.131   0.175     0.132     0.15

Diabetes            0.046   0.059     0,041     0.05

Back                 0.27    0.23      0.19     0.23

Migraine            0.139   0.113      0.12    0.122

Other chronic      0.1211  0.1112    0.0912    0.108

Active               0.43    0.53     0.517    0.498

Binge               0.357   0.311     0.258    0.308


Our outcome of interest, number of hospital days, is a count variable with two main features. First, there is over dispersion in the data, that is, the variance is larger than the mean. Second, there is a large number of zeros in the number of nights at hospitals. We will employ two specific count estimators based on the negative binomial distribution, that is, the negative binomial (NB) model and the zero inflated negative binomial (ZINB) model that can accommodate these features. (9) They are described in detail below.

The NB model is an extension of the basic Poisson model which is often used to model count variables. While assuming a Poisson distribution for the count, like the basic Poisson model, the NB model allows the rate parameter of the Poisson distribution to be random and follow a separate distribution. This separate distribution can accommodate the over dispersion feature. If one thinks of the source of this feature as unobserved heterogeneity, then the NB model is a mixture distribution model that assumes the Poisson distribution for the count variable and a separate distribution for the unobserved heterogeneity.

More formally, the NB model is defined by the distribution: P(y) =[e.sup.-[lambda]][[lambda].sup.y]/y!, with the rate parameter [lambda] = [micro]v, where [micro] is a deterministic function of X. exp(X'[beta]) and v is assumed to follow a Gamma distribution.

The conditional mean and variance of the NB distribution are given by

Mean = E[y | [micro], [alpha]] = [micro] = exp(X'[beta])

Variance = V [y | [micro], [alpha]] = [micro] + [alpha]g([micro])

The NB model can be estimated by maximum likelihood. As the negative binomial distribution model has exponential conditional mean, its coefficient estimate can be interpreted as a semi-elasticity (Cameron and Trivedi 2009). Henceforth, in our discussion, when we use the word "cost" we intend it in the relative sense: a coefficient of 0.3, for example, on a particular variable means that a unit increase in that variable imposes an additional 30% cost.

The ZINB model has two parts, one where members always have zero counts and one where members have positive counts. The likelihood of being in either regime is estimated using a logit specification, while the counts in the second regime are estimated using a negative binomial specification. While both models can accommodate excess zeros, the ZINB model differs from the NB model in that it allows for the process of generating the zeros to be different from the process of determining the positive values.

Pr(Y =O | X) = [f.sub.1](0 | [[alpha].sub.1]) + (1 - [f.sub.1](0 | [[alpha].sub.1]))[f.sub.2](0 | [[alpha].sub.2])

Pr(Y = y > 0 | X) = (1 - [f.sub.1](0 | [[alpha].sub.1]))[f.sub.2](0 | [[alpha].sub.2])

where [f.sub.1](.| [[theta].sub.1]) is a logit/probit model and [f.sub.2](.| [[theta].sub.1]) is a Poisson or NB model.

Given that we obtained similar results with each of these two estimators, we will focus on reporting the negative binomial model results because the coefficient estimates from the NB model are more easily interpretable. However, we also present the results from our ZINB models subsequently.


A. Negative Binomial Model Results

Initial estimates of excess hospitalization costs are contained in Table 3. We group individuals into three categories from a smoking standpoint. The reference group is "never smoked," that is, not having smoked 100 cigarettes in the lifetime, and the two remaining groups are current and former smokers. This table contains results for two age groups for comparison purposes. The first two columns contain the results for the working-age group, the second two columns for the entire population aged 20 and above. All specifications also contain a set of fixed effects for the province of residence (of which there are 10) and year. The estimates are slightly sensitive to the age band chosen, and this sensitivity is evident where former smokers are concerned.
                                   TABLE 3

Smoking Status and Number of Hospitalized Days, Negative Binomial Model

                        (1)        (2)        (3)        (4)

Covariates           Age 25-65  Age 25-65     Age        Age
                                             20-80+     20-80+

Current smoker        0.286***   0.171***   0.303***   0.199***

                      (0.0687)   (0.0572)   (0.0598)   (0.0441)

Former smoker           0.149*     0.0655   0.201***    0.135**

                      (0.0800)   (0.0695)   (0.0645)   (0.0633)

Obese                            0.266***              0.253***

                                 (0.0214)              (0.0193)

Chronic dummy                    0.950***              0.935***

                                 (0.0263)             (0.03 11)

Active               -0.289***  -0.236***  -0.329***  -0.285***

                      (0.0302)   (0.0352)   (0.0318)   (0.0356)

Student               -0.316**   -0.319**  -0.428***  -0.414***

                       (0.134)    (0.135)   (0.0912)   (0.0876)

Male                 -0.592***  -0.551***  -0.465***  -0.418***

                      (0.0780)   (0.0736)   (0.0601)   (0.0549)

Age 35-44            -0.279***  -0.440***     0.0126    0.00562

                      (0.0229)   (0.0185)   (0.0641)   (0.0721)

Age 45-54            -0.148***  -0.428***  -0.301***  -0.467***

                      (0.0394)   (0.0588)   (0.0494)   (0.0584)

Age 55-64             0.358***    0.00129  -0.193***  -0.479***

                      (0.0387)   (0.0364)   (0.0411)   (0.0605)

Age 65-74                                   0.286***    -0.0791

                                            (0.0519)   (0.0513)

Age 75-79                                   0.721***   0.336***

                                            (0.0626)   (0.0649)

Age 80+                                     1.066***   0.666***

                                            (0.0321)   (0.0320)

Married               0.141***   0.160***   0.203***   0.219***

                      (0.0494)   (0.0591)   (0.0356)   (0.0439)

Hhsize               -0.0298**    -0.0118  -0.0361**    -0.0217

                      (0.0132)   (0.0163)   (0.0158)   (0.0171)

Secondary school     -0.302***  -0.204***  -0.222***  -0.148***

                      (0.0244)   (0.0249)   (0.0233)   (0.0267)

Some post-secondary  -0.303***  -0.233***  -0.269***  -0.231***

                      (0.0544)   (0.0412)   (0.0580)   (0.0426)

Post-secondary       -0.337***  -0.228***  -0.275***  -0.191***

                      (0.0256)   (0.0513)   (0.0146)   (0.0285)

Immigrant            -0.209***  -0.191***  -0.170***  -0.143***

                      (0.0324)   (0.0293)   (0.0259)   (0.0233)

French speaking       0.147***   0.142***   0.129***   0.112***

                      (0.0399)   (0.0355)   (0.0343)  (0.031 1)

Constant             -0.490***  -1.127***  -0.638***  -1.214***

                      (0.0401)   (0.0659)   (0.0820)   (0.0977)

Observations           222.649    222.611    319.833    319.783

Notes: Province fixed effects and yearly fixed effects are included;
province-clustered standard errors in parentheses.
*p < .1. **p < .05. ***p < .01.

Column 1 indicates that current smokers have approximately twice the excess cost of former smokers. The continuing high cost of former smokers is not unusual in the literature. For example, Cremieux et al. (2010) find that the coefficients on these two variables are almost identical. The coefficient 0.286 for current smokers indicates that, relative to never smokers, they stay almost 30% longer in hospital per year. The costs based on the full age sample are slightly higher for both current and former smokers.

When a measure of obesity, and a covariate indicating if the individual suffers from some chronic condition or not (columns 2 and 4) are included, the coefficients on current and former smokers drop; this illustrates the importance of the whole spectrum of health conditions on hospitalization. In the wider age group the coefficient on former smokers remains statistically significant, but in the prime age group the former smoker category loses its significance.

The activity dummy takes on value 1 if the respondent is physically active (either moderate or high). The significance on the activity variables indicates that lifestyle is important, independent of obesity status. That is, inactivity without obesity is detrimental per se, and activity even in the presence of obesity is cost reducing. (10)

The socio-demographic variables carry the expected signs: Older individuals use hospitals more than younger individuals; those with any level of post-secondary education use hospitals less than those who have not completed secondary education; males use hospitals less frequently than females--largely because of births. The greater use by married individuals likewise reflects births. Immigrants use hospitals slightly less than native-born individuals. Language is also significant though minor in its impact; its impact may reflect cultural practices rather than indicate that those who speak a different language are less healthy.

The initial set of results in Table 3 has no time dimension to it: former smokers are treated as one group, regardless of for how many years they smoked, or the time period since they quit. Thus, the "former smoker" coefficient in Table 3 is an average impact regardless of the history of those individuals. Hence, if costs following quitting are initially high and subsequently decline, the coefficient yields little information on the dynamics of costs. A number of studies (as detailed by the 10M 2009) have documented a variety of patterns when time-since-quitting is taken into consideration. For example, Kahende et al. (2009) find that hospitalizations following quits have a U shape: they decline slightly in the years following a quit decision but appear to increase again 10 years after the quit deci-sion--despite controls for age. Fishman et al. (2006) provide a second example. They studied the medical records of a group of insured individuals in the United States for a 4-year period and concluded that the medical costs associated with individuals who quit actually begin to rise in the year preceding the transition from smoker to nonsmoker status. They also cite other work indicating that deterioration in health status is a good predictor of quits. Their costs appear to diminish rapidly 'after cessation (1 year).

We thus disaggregate the former smokers by number of years since quitting, in addition to recognizing whether current smokers are daily or occasional, and whether they were daily or .occasional before quitting. The results from this disaggregation for the prime age group are presented in column 1 of Table 4. They indicate that daily smokers are much more costly than occasional smokers. (11) The pattern for former smokers becomes very clearly defined, and is somewhat surprising. A recent quitter (one who quit within the preceding 2 years) costs more than twice what a current daily smoker costs (0.57 vs. 0.23). A quitter of more than 2 years but less than or equal to 5 years still has a cost that approximates that of current daily smokers. However, once we hit a 6-year lag the cost declines substantially and significantly. The coefficient on the more-than-6-year quitter is no longer significantly different from zero. These coefficient estimates suggest that, from an employment perspective, a former smoker may be less costly over her whole life cycle, but for .a period of 6 years she will be at least as costly as a current smoker. Few employers have horizons longer than this. Interestingly, the coefficient associated with former occasional smokers who quit is negative. This may reflect their personality type: a person who can smoke just occasionally without becoming habituated, and then be able to quit successfully, has likely more control over his or her life than an average individual.
                     TABLE 4

  Health Care Costs and Breakdown of Smoking
  Status, Age 25-65, Negative Binomial Model

Covariates          (1)        (2)        (3)

Current daily    0.226***   0.236***   0.233***
                 (0.0644)   (0.0607)   (0.0808)
Current occas.    0.00880     0.0225    -0.0227
                 (0.0537)   (0.0547)   (0.0478)
Daily former     0.570***   0.579***   0.475***
2 years          (0.1 12)    (0.113)   (0.0747)
Daily former      0.192**     0.200"     0.191*
3_5 years        (0.0966)   (0.0940)    (0.101)
Daily former       0.0292     0.0374    0.00373
6 year over      (0.0759)   (0.0738)   (0.0791)
Occas. former    -0.143**   -0.135**   -0.129**
                 (0.0623)   (0.0590)   (0.0512)
Obese            0.264***   0.263***   0.201***
                 (0.0220)   (0.0223)   (0.0273)
Active          -0.242***  -0.243***  -0.215***
                 (0.0336)   (0.0333)   (0.0324)
Binge                       -0.115**
Chronic          0.935***   0.934***
                 (0.0233)   (0.0242)
Asthma                                 0.473***
Heart                                  1.755***
Hyper                                  0.447***
Diabetes                               0.733***
Back pain                              0.439***
Migraine                               0.328***
Other chronic                          1.043***
Constant        -1.195***  -1.184***  -1.149***
                 (0.0554)   (0.0591)   (0.0468)
Observations      222.611   222.61 1    221.519

Notes: Demographic controls, province fixed effects, and yearly fixed
effects are included; province-clustered standard errors in
*p < .1, **p < .05, **p < .01.

In addition to the obesity and activity variables this regression includes a further behavioral marker--the presence or absence of binge drinking--defined as consuming five or more drinks on a single occasion. Surprisingly this turns up with a small negative sign and is marginally significant. Obviously it would be imprudent to advocate such activity. It is possible that reporting is selective and only those who are in reasonable health are sufficiently confident to report their behavior. The obesity coefficient remains significant and now exceeds the cost associated with being a daily smoker. Again, this coefficient is an average for all individuals having a BMI [greater than or equal to] 30. Were we to distinguish between individuals with differing BMI levels, this coefficient too would likely manifest different impacts depending on the degree of obesity. (12)

In Table 4 we also break out a subsample of the chronic conditions variables in order to examine if some are more costly than others. We single out six such conditions and group the remaining conditions into the single "other chronic .conditions" variable. The coefficient on this other chronic conditions variable remains similar in magnitude and significance to the overall chronic variable in Table 3. Each of the six individual conditions is highly significant and there is a wide disparity in their impact. For example, migraine and back pain are about one quarter as costly as a heart condition and about one half as costly as diabetes.

These differences are important from the standpoint or hiring, if an employer's objective is to minimize workdays lost and insurance costs paid to an insurer on behalf of the employees. Up-front screening for a large number of chronic conditions by an employer may be problematic and costly--partly because a potential employee may not be aware .of one or more of his or her own conditions. Our primary objective here is to illustrate that being a smoker is but one observable medical condition among many that leads to elevated employee costs; furthermore costs vary dramatically by condition.

To gain further insight into past behavior of smokers we next examine the impact of the duration of smoking for those who quit. Table 5 reports the results for three groups of former smokers--those who smoked for less than 10 years, for between 10. and 20 years, and for more than 20 years. We continue to focus on those of working age. 25-65, (13) and maintain the distinction within each regression between smokers who quit recently and further in the past. This breakdown adds considerably to our understanding. It is the group of long-time smokers that is the most costly, particularly if they are recent quitters. Among those who have quit in the most recent 2 years, 20-year-plus smokers are more than twice as costly as 10-20 year smokers and almost four times as costly as those who smoked for less than 10 years. Furthermore, as we move beyond the recent quitters to those who quit between three and 5 years ago, or more than 6 years ago, it is the long-time smokers who continue to be significantly more costly than never smokers. (14)
                                 TABLE 5

Health Care Costs and the Length of Smoking Before Quit, Age 25-65,
Negative Binomial Model

Covariates                    (1)         (2)          (3)
                            Column      Sample =     Sample =
                          3 of Table  Nonsmokers    Nonsmokers
                            4 Above    + Current     + Current
                          Reproduced   Smokers +     Smokers +
                                        Former        Former
                                      Smokers Who   Smokers Who
                                      Smoked for    Smoked for
                                      < 10 Years    Between 10
                                                    and 20 Years

Current daily               0.233***     0.266***      0.208***
                            (0.0808)     (0.0748)      (0.0758)

Current occas.               -0.0227    -0.000111       -0.0372
                            (0.0478)     (0.0441)      (0.0425)

Daily former 2 years        0.475***     0.224***      0.244***
                            (0.0747)     (0.0541)      (0.0917)

Daily former 3_5 years        0.191*      -0.277*        0.0691
                             (0.101)      (0.153)       (0.108)

Daily former 6 year over     0.00373    -0.170***       -0.0339
                            (0.0791)     (0.0643)       (0.153)

Occas. former               -0.129**      -0.0704     -0.634***
                            (0.0512)     (0.0574)      (0.0902)

Obese                       0.201***     0.256***      0.236***
                            (0.0273)     (0.0316)      (0.0413)

Active                     -0.215***    -0.200***     -0.182***
                            (0.0324)     (0.0299)      (0.0255)

Asthma                      0.473***     0.471***      0.529***
                            (0.0797)     (0.0589)      (0.0672)

Heart                       1.755***     1.660***      1.730***
                            (0.0567)     (0.0616)      (0.0619)

Hyper                       0.447***     0.429***      0.411***
                            (0.0461)     (0.0385)      (0.0480)

Diabetes                    0.733***     0.773***      0.754***
                            (0.0599)     (0.0351)      (0.0271)

Back                        0.439***     0.483***      0.468***
                            (0.0295)     (0.0138)      (0.0296)

Migraine                    0.328***     0.262***      0.342***
                            (0.0499)     (0.0313)      (0.0423)

Other chronic               1.043***     1.099***      1.044***
                            (0.0539)     (0.0689)      (0.0770)

Constant                   -1.149***    -1.168***     -1.171***
                            (0.0468)      (0.122)       (0.104)

Observations                 221.519      167.227       149,024

Covariates                     (4)
                            Sample =
                            + Current
                            Smokers +
                           Smokers Who
                            Smoked for
                          [greater than
                          or equal to]
                            20 Years

Current daily                  0.208***

Current occas.                  -0.0369

Daily former 2 years           0.755***

Daily former 3_5 years         0.519***

Daily former 6 year over        0.191**

Occas. former                   -0.0938

Obese                          0.236***

Active                        -0.180***

Asthma                         0.551***

Heart                          1.710***

Hyper                          0.450***

Diabetes                       0.686***

Back                           0.485***

Migraine                       0.282***

Other chronic                  1.015***

Constant                      -1.136***

Observations                    157.600

Notes: Demographic controls, province fixed effects, and yearly fixed
effects are included; province-clustered standard errors in
*p < .1. **p < .05. ***p < .01.

An implication of why recent quitters--particularly those of long duration--are so costly is that prior to quitting they were a subset of then current smokers with health conditions that prompted them to quit. (15) Fishman et al. (2006) suggest that the acuteness of these other conditions may have been aggravated by smoking and subsequently been moderated by the quit decision. The benefits of quitting for such individuals evidently require a significant period of time to materialize. Our results are consistent with the possibility that smoking may exacerbate other chronic conditions and that quitting abates these other conditions over time. The same holds for individuals who were formerly obese--their dietary habits and sedentary lifestyle contributed to a buildup of cholesterol which in turn led to a cardiac condition or diabetes.The severity of the latter conditions can diminish with a BMJ reduction.

The critical focus of our analysis is upon observable characteristics that determine which potential employees will be most costly to employers. It matters little to employers whether smoking or obesity may have contributed more to a current diabetes or cardiac condition. Nonetheless, the results in Table 5 indicate that smoking is not a significant factor in the development of most chronic conditions: If smoking were indeed a causative factor in generating chronic conditions, then individuals who smoke for longer periods of time should have larger coefficients on the chronic conditions variables. In particular, individuals who smoke or smoked for more than 20 years provide the toxins in tobacco a longer period of time to generate chronic conditions. But this is not revealed in the pattern of coefficients on the chronic variables as we move from one duration of smoking to another in Table 5. The chronic coefficients are remarkably stable as we move from column 2 to column 4. This is strong evidence that the smoking effect on health is not being camouflaged by its impact on chronic conditions.

As a final comment here, the foregoing discussion implies that the higher cost coefficient value for recent quitters does not imply that those quitters would be, in effect, lower cost individuals by continuing to smoke.

B. Zero Inflated Negative Binomial Results

In addition to the negative binomial model, we also estimated the zero-inflated negative binomial (ZINB) model. The "number of nights at hospital" as an outcome in the NB model analysis can be thought of as the product of two parameters: the probability of being hospitalized and, conditional on being hospitalized, the actual number of nights in hospital. As noted earlier, the NB model assumes that the processes of generating these two outcomes are the same. The ZINB model relaxes this assumption.

The results are presented in Tables 6 and 7. Table 6 (equivalent to Table .4 for the NB model) indicates that current .smokers who smoke daily incur a higher likelihood of having medical problems than nonsmokers, both in terms of risk of being hospitalized as well as the length of stay in hospital once hospitalized. Current occasional smokers have less risk of being hospitalized than the current daily smokers, although the coefficients are of the same direction as the effect for current daily smokers. For quitters, within 2 years of their quitting, the risks of being hospitalized and of a lengthy stay in hospital are .still very high (even higher than that of current smokers). These risks of adverse health outcomes, however, drop after the third year of quitting, and continue to drop after the sixth year, reflected in the smaller coefficients (in absolute value) than the coefficients on the 2-year quit dummy variable.

Results for other covariates show that being Obese or having .a chronic condition (in general) carries a higher risk of being hospitalized and staying longer in hospital (columns 1 and 2). Back pain, migraine, and hypertension carry higher risk of hospitalization than no chronic condition at all, but there is little difference in terms of hospital stay length. Meanwhile, asthma, diabetes, and .cardiac diseases display both higher risk or hospitalization and require a longer stay than no condition at all. Further, asthma, diabetes, and cardiac conditions are more .costly than smoking, captured by their larger coefficient -estimates in both zero count and positive count regressions.
                            TABLE 6

Health Care Costs and Breakdown of Smoking Status, Age 25-65.
Zero-Inflated Negative Binomial Model

Covariates     (1) Zero Count  (2) Positive  (3) Zero Count
               Part (Prob. of   Count Part   Part (Prob. of
                 Not Being         (# of       Not Being
               Hospitalized)      Nights)     Hospitalized)

Current           -0.143***       0.0583*       -0.126***
Daily              (0.0271)      (0.0314)        (0.0278)

Current            -0.0812*       0.01 15         -0.0666
occas.             (0.0471)      (0.0536)        (0.0479)

Daily former      -0.449***      0.218***       -0.418***
2 years
                   (0.0388)      (0.0429)        (0.0396)

Daily former        -0.0814       0.123**         -0.0329
3_5 years          (0.0536)      (0.0612)        (0.0549)

Daily former       -0.0576*       -0.0324         -0.0334
6 year over        (0.0296)      (0.0345)        (0.0304)

Occas.               0.0407     -0.0876**          0.0328
Former             (0.0312)      (0.0360)        (0.0318)

Obese             -0.210***       0.0428*       -0.127***
                   (0.0223)      (0.0250)        (0.0235)

Chronic           -0.640***      0.288***
                   (0.0220)      (0.0266)

Asthma                                          -0.381***

Heart                                           -1.481***

Hyper                                           -0.415***

Diabetes                                        -0.468***

Back                                            -0.299***

Migraine                                        -0.277***

Other chronic                                   -0.494***

Active             0.160***     -0.133***        0.138***
                   (0.0193)      (0.0221)        (0.0198)

Constant           2.051***      1.029***        2.010***
                   (0.0721)      (0.0837)        (0.0734)

Observations        222.611      222,61 1         221,519

Covariates     (4) Positive
                Count Part
               (# of Nights)

Current           0.0607*
Daily            (0.0315)

Current           0.00633
occas.           (0.0537)

Daily former     0.212***
2 years

Daily former      0.123**
3_5 years        (0.0614)

Daily former      -0.0324
6 year over      (0.0345)

Occas.          -0.0879**
Former           (0.0361)

Obese              0.0316


Asthma           0.153***

Heart            0.358***

Hyper             0.0501*

Diabetes         0.205***

Back            0.0859***

Migraine         0.0607**

Other chronic    0.419***

Active          -0.126***

Constant         1.053***

Observations      221.519

Notes: Demographic controls, province fixed effects, and yearly fixed
effects are included; province-clustered standard errors in
*p < .1, **P < .05, ***p < .01.

                       TABLE 8

Health Care Costs and Breakdown of Smoking Status.
      Age 25-40. Negative Binomial Model

Covariates         (1)        (2)        (3)
Current daily   0.224****   0.245***   0.226***
                 (0.0685)   (0.0671)   (0.0825)
Current occas.    -0.0412    -0.0213    -0.0648
                  (0.136)    (0.136)    (0.127)
Daily former     0.225***   0.240***   0.237***
2 years          (0.0605)   (0.0639)   (0.0654)
Daily former     -0.235**    -0.219*    -0.240*
3_5 years         (0.125)    (0.132)    (0.128)
Daily former      -0.102*   -0.0892*  -0.0942**
6 year over      (0.0544)   (0.0521)   (0.0412)
Occas. former    -0.105**   -0.0904*  -0.104***
                 (0.0477)   (0.0520)   (0.0416)
Obese           0.275****   0.276***   0.284***
                 (0.0384)   (0.0385)   (0.0298)
Active           -0.111**  -0.108***    -0.109*
                 (0.0557)   (0.0547)   (0.0632)
Binge                      -0.148***
Chronic          0.691***   0.690***
                 (0.0408)   (0.0427)
Asthma                                 0.387***
Heart                                  1.439***
Hyper                                  0.425***
Diabetes                               1.065***
Back pain                              0.429***
Migraine                                  0.174
                                       (0.1 17)
Other chronic                          0.990***
Constant        -1.375***  -1.354***  -1.410***
                 (0.0515)   (0.0525)   (0.0584)
Observations       79.937     79.937     79.645

Notes: Demographic controls, province fixed effects and yearly fixed
effects are included; province-clustered standard errors in
*P < .1, **p < .05, ***p < .01.

The results in Table 7 (equivalent to Table 5 for the NB model) define the long-term health consequences of long-time smoking. Among those who quit within 2 years, those who smoked for a longer time will have higher risk of being hospitalized and, if admitted to hospital, will spend more nights than those who smoked for a shorter time period. Further, those who smoked for more than 20 years have significantly higher risk of being hospitalized, even if they quit over 6 years .ago.

Overall, the ZINB models are consistent with the results from the NB models. They suggest that smoking affects health care costs in two ways: raising the risk of being hospitalized for the current smoker and also for recent quitters (i.e., a lagged effect), and increasing the number of nights spent in hospitals if they are hospitalized. The ZINB models also indicate that individuals with some chronic conditions have even higher risk of hospitalization and face longer stays than smokers.

Our final robustness check deals with the possibility that truncation (by death) may bias the samples. Consider the following: two individuals die in the year preceding_ the survey, one from a smoking-related condition (e.g., myocardial infarction), and the other from a stroke, caused by obesity or high blood pressure or a combination of the two. Clearly neither of these individuals can answer the survey questions on how many days they spent in hospital the year prior to the survey. Yet if one cause of death typically involves more days hospitalized than the other, then our results could underestimate the number of days associated with the longer palliative period. To test for such a possibility we ran the models for age groups where mortality is low--for those under the age of 40 (the median age of stroke and myocardial infarction victims is in the neighborhood of 70 in developed economies, Pell et al. 2008). The estimates obtained and shown in *Tables 8 and 9 were very similar to those reported for the complete sample, suggesting that this potential source of bias is not present in our results.

The results in Tables 8 and 9 also shed light on the source of high health care costs of former smokers. In particular, the coefficients on the recent quitters (within 2 years) are positive and statistically significant. Because chronic conditions are less prevalent among smokers in this young age group, these high costs for these recent quitters are likely to be smoking related. This suggests that it may be inefficient for employers to single out only current smokers for hiring discrimination, while recent quitters also impose high costs on employers due to their past smoking.
                              TABLE 9

    Health Care Costs and the Length of Smoking Before Quit.
              Age 25-40, Negative Binomial Model

Covariates          (1) Sample =      (3) Sample =      (5) Sample =
                    Nonsmokers +      Nonsmokers +      Nonsmokers +
                  Current Smokers   Current Smokers   Current Smokers
                  + Former Smokers  + Former Smokers  + Former Smokers
                  Who Smoked for<    Who Smoked for    Who Smoked for
                      10 Years       Between 10 and     [greater than
                                        20 Years       or equal to]20

Current daily          0.235***           0.191**           0.197**
                       (0.0802)          (0.0837)          (0.0855)

Current occas.          -0.0664           -0.0892           -0.0918
                        (0.128)          (0.1 18)           (0.120)

Daily former 2         0.213***            0.273*            -1.032
years                  (0.0374)           (0.161)           (1.387)

Daily former 3_5      -0.427***            0.1 10
years                   (0.148)           (0.165)

Daily former 6         -0.126**             0.473
year over              (0.0551)           (0.326)

Occas. former          -0.0735*         -0.832***         -19.95***
                       (0.0434)           (0.142)           (0.855)

Obese                  0.303***          0.336***          0.356***
                       (0.0381)          (0.0548)          (0.0628)

Active                 -0.0872*           -0.0646           -0.0387
                       (0.0522)          (0.0769)          (0.0660)

Asthma                  0412***          0.501***          0.531***
                       (0.0785)          (0.0643)          (0.0685)

Heart                  1.439***          1.416***          1.375***
                        (0.133)           (0.125)          (0.0955)

Hyper                  0.466***          0.531***          0.579***
                        (0.105)          (0.0972)          (0.0965)

Diabetes               1.102***          1.050***         1.1 00***
                       (0.0951)           (0.102)          (0.1 14)

Back pain              0.410***          0.480***          0.469***
                       (0.0336)          (0.0578)          (0.0493)

Migraine                  0.155             0.130             0.106
                        (0.107)           (0.134)           (0.131)

Other chronic          1.001***          0.986***          0.994***
                       (0.0623)          (0.0750)          (0.0856)

Constant              -1.415***          1.398***         -1.379***
                       (0.0597)          (0.0907)           (0.123)

Observations             76.061            55,103            51.605

Nates: Demographic controls, province fixed effects, and yearly fixed
effects are included; province-clustered standard errors in
*p < .1, **p < .05, ***p < .01.


This paper was motivated by the legality of discriminatory hiring against smokers in 21 American states. Our value added to the small literature that characterizes this growing area of investigation takes the form of examining whether such discrimination against current smokers is the most effective cost-reducing strategy for an employer, given that many different morbidities and habits reduce on-the-job productivity. Our measure of cost or inefficiency is the number of days that individuals spend in hospital each year.

Three major results emerge from analyzing several very large recent health surveys: The 'first is that the daily smokers are no more costly to employers than individuals who are obese or who suffer from a range of health conditions. In many cases smokers are less costly.

The second result is that individuals in the prime working years who are nondaily (occasional) smokers are not significantly more costly than individuals who are never smokers. Hence there is little or no value resulting from discriminating against this type of smoker. It should be emphasized that a substantial proportion of individuals termed "smokers" fall into this category; the CCHS surveys indicate that the proportion is one-fifth.

Our third result relates to former smokers. We find repeatedly that former smokers who are recent quitters (2 years or less) are more costly than current smokers on average. This result may be attributable to the fact that individuals whose health is particularly poor prior to quitting are overrepresented in the actual quitter group. Nonetheless. this result states that employers who fail to discriminate against recent former smokers, particularly those who smoked for a long duration, are not acting efficiently. Our results indicate that recent quitters who smoked for more than 20 years are significantly and substantially more costly than recent quitters who smoked for shorter periods of time. We find that individuals who quit in our data set need a- period of about 6 years before they approach the behavior of never smokers. Few employers have horizons this long.

To conclude: it is well accepted that discriminating against smokers impacts most strongly those individuals from lower socio-economic groups. Our results indicate furthermore that a simple hiring rule which takes the form of discrimination against current smokers is a very poor one from an efficiency standpoint.

Numerous employers in over 20 U.S. states currently discriminate legally against smokers in their hiring policies. We analyze the cost of being a smoker, measured in annual hospital days, and compare this with the cost of being a former smoker, the cost of being obese, and the cost of a variety of other medical conditions, relative to the cost of being a never smoker, using three large recent surveys each having in excess of one hundred thousand observations. The paper also explores the cost of former smokers as determined by the number of years since quitting. Smokers as a whole are not found to be the most costly employees. Furthermore, health costs vary dramatically among smokers of different duration and intensity. As a consequence, our results question the efficiency of such discrimination. (JEL HO, 118, J71. J7)


CCHS: Canadian Community Health Surveys

IOM: Institute of Medicine

NB: Negative Binomial

SHS: Second-Hand Smoke

ZINB: Zero Inflated Negative Binomial

Irvine: Department of Economics. Concordia University, 1455 de Maisonneuve Blvd. West, suite H 1155. Montreal. Quebec. Canada H3G 1M8. Phone (1) 514 848 2424 ext. 3909. Fax 514 848 4563, E-mail i an

Nguyen: Program in Health Services and System Research, Duke-NUS Graduate Medical School Singapore, 8 College Road. Singapore 169857. Phone (65) 6601 3576. Fax (65) 6534 8632, E-mail


Banerjee, A., and S. Mullainathan. "The Shape of Temptation: Implications for the Economic Lives of the Poor. "NBER. 2010.

Becker, G., and M. K. Murphy. "A Theory of Rational Addiction." Journal of Political Economy, 96(4), 1988, 675-700.

Bernheim. D., and A. Rangel. "From Neuroscience to Public -Policy: A New Economic View of Addiction." Swedish Economic Policy Review, 12. 2005. 95-144.

Bhattacharya, J., and M. K. Bundorf. "The Incidence of the Healthcare Costs of Obesity." Journal of Health Economics. 28(3). 2009, 649-58.

Bhattacharya, J., .and N. Sood. "Who Pays for Obesity?" Journal of Economic Perspectives, 25(1), 2011, 139-58.

Burkhauser, R., and J. Cawley. "The Value of More Accurate Measures of Fatness and Obesity in Social Science Research." Journal of Health Economics, 27(2), 2007, 519-29.

Cameron, C., and P. Trevedi. Micmcconometrics: Methods and Applications. New York: Cambridge University Press, 2005.

--. Microeconometrics Using Stata. College Station, TX: Stata Press, 2009.

Cawley. J. "The Impact of Obesity on Wages." Journal of Human Resources. 39(2). 2004. 451-74.

Cleveland Clinic. 2007.

Cremieux. P.-Y., L. Pinheiro. M. Ginn, and M. Cloutier. Impact Direct du Tabagisme sur le Systeme de Sante au Quebec. [much greater than] Groupe D'Analyse, pour le compte de la Coalition quebecoise pour le controle du tahac, 2010.

Critchley, J., and S. Capewell. "Mortality Risk Reduction Associated with Smoking Cessation in Patients with Coronary Heart Disease: A Systematic Review." Journal of the American Medical Association, 290(1). 2003, 86-97.

Della Vigna. S. "Psychology and Economics: Evidence from the Field." Journal of Economic Literature, 47(2), 2009. 3.15-72.

Fishman, P., E. Thompson, E. Merikle, and S. Curry. "Changes in Health Care Costs Before and After Smoking Cessation." Nicotine Tobacco Research.. 8(3), 2006. 393-401.

Florescu, A., R. Ferrence, T. Einarson, P. Selby, 0. Soldin and G. Koren. "Methods for Quantification of Exposure to Cigarette Smoking and Environmental Tobacco Smoke: Focus on Developmental Toxicology." Therapeutic Drug Monitor, 31(1). 2009. 14-30.

Gastaldo. E. "Fort Worth May Refuse to Hire Smokers." 2012.

Gruber, J., B. Koszegi. "Tax Incidence When Individuals Are Time Inconsistent: The Case of Cigarette Taxes." Journal of Public Economics. 88(9-10), 2004, 1959-87.

Gul, F., and W. Pessendorfer. "Temptation and Self Control." Econometrica. 69(6). 2001, 1403-35.

Institute of Medicine. Second-Hand Smoke Exposure and Cardiovascular Effects: Making Sense of the Evidence. Washington. DC: National Academies Press, 2009.

Jones, A. M. 2010. "Models for Health Care." Health. Econometrics. and Data Group (HEDG) Working Papers 10/01, HEDG, University of York.

Kahende. J., B. Adhikari, E. Maurice. V. Rock. and A Malarcher. "Disparities in Health Care Utilization by Smoking Status--NHANES 1999-2004." International Jounzal of Environmental Research in Public Health. 6(3). 2009, 1095-1.06.

McDaniel. R. and R. Malone. "Businesses' Voluntary Pro-health Policies: A Review and Research Agenda." Tobacco Control. 21, 2012, 66-72.

Mullahy, J. "Econometric Modeling of Health Care Costs and Expenditures: A Survey of Analytical Issues and Related Policy Considerations." Medical Care. 47,(7_Supplement_1), 2009, S104-08.

O'Donoghue. T., and M. Rabin. "Risky Behavior Among Youths: Some Issues from Behavioral Economics." in Risky Behavior Among Youths: An Economic Analysis, edited by J. Gruber. Chicago: University of Chicago Press, 2001.

Pell. J., S. Haw, S. Cobbe. D. Newby, A. Pell. C. Fischbacher. A. McConnachie. S. Pringle. D. Murdoch, F. Dunn. K. Oldroyd. P. MacIntyre. B. O'Rourke. and W. Borland. "Smoke-free Legislation and Hospitalizations for Acute Coronary Syndrome." New England Journal of Medicine, 359, 2008. 482-91.

RT. "Florida City Passes Ban on Hiring Smokers," October 3, 2012.

Shetty, K., T. DeLeire. C. White. and J. Battacharya. "Changes in Hospitalization and Mortality Rates Following Smoking Bans." Journal of Policy Analysis and Management. 30(I). 2010, 6-28.

Siegel. M. "Challenge Issued to Smoker-Free Hospitals: Put Lip or Shut Up." 2011.

Sulzberger, A. G. "Hospitals Shift Smoking Bans to Smoker Ban." New York Times. February 10, 2011.

Voigt, Kristin. "Nonsmoker and 'Non-nicotine' Hiring Policies: The Implications of Employment Restrictions for Tobacco Control." American .Journal. of Public Health, 102, 2012, 2013--18.

Zamora, K. "Baptist Health jobs Not Open to Smokers." San Antonio Post. October. 29. 2012.

(1.) In contrast. some nonprofit organizations devoted to the improvement of health and the elimination of smoking. such as the American Legacy Foundation, do not favor these practices.

(2.) The methodology underlying this vein of research is similar to ours--it tends to focus on hospital admissions, generally for a range of cardiac disorders. The efficacy of prohibiting smoking in public places is then determined by comparing hospital admissions pre and post ban. An interesting recent overview and analysis is contained in Shetty et al. (2010).

(3.) It appears that antismoker policies tend to apply to new potential employees only and that existing employees are given an exemption. This is the case at Cleveland and also at Baylor (Gastaldo 2012).

(4.) One reason is lime inconsistency on the part of the consumer. Gruber and Koszegi (2004) propose that smokers have high near-term discount rates and lower discount rates further into the future, at least from today's perspective. In this "two selves" world, if the lower discount rate is the appropriate one, then government policy can induce individuals to act in a manner that is consistent with the lower discount rate through the appropriate calibrating of its polic instruments. A second reason is that individuals may suffer from what O'Donohue and Rabin (2001) call projection bias--individuals may not know the form of their future utility. Third. Bernheim and Rangel! (2005) propose that the brain can make errors when subject to specific types of cue. Hence, regulation of images that stifles such cues could improve the decision-making and well-being of individuals who have a tendency to consume products such as tobacco. Finally. Gul and Pessendorfer (2001) and Banerjee and Mullainathan (2010) propose that individuals may have more than one decision mechanism in their brain: In addition to a longer-term "rational" calculus, individuals may be subject to temptation. While actual choices reflect the interplay of these two mechanisms, if we deem the rational mechanism to be superior, then policies designed to suppress temptations by restricting the menu of choices can improve well-being. This literature is reviewed extensively by Della Vigna (2009).

(5.) We are grateful to a referee for drawing our attention to this issue.

(6.) A referee has pointed out to us that this system of insurance differs from what existed in the United States at the time .of our surveys. The Canadian universal public coverage poses a moral hazard to every individual in the economy, whereas the primarily private coverage in the United States carries differential moral hazard. Furthermore, we cannot say if the cost pattern that is associated with the various morbidities we examine is invariant to the replacement of one system of coverage with another. Nonetheless, if the "Obamacare" legislation governing compulsory insurance survives (and the odds of that happening since the November 2012 election have increased considerably). the U.S. system will mimic the Canadian system more closely. in that most of the U.S. population will be covered and therefore virtually every individual will suffer from moral hazard--similar, but not identical to, Canada.

(7._ The number of hospitalization days is top constrained to be 31 in the CCHS surveys.

(8.) Asthma. arthritis. back problems, high blood pressure. migraine, chronic bronchitis, emphysema. COPD, diabetes, heart disease, cancer. ulcers. effects of a stroke, incontinence. bowel disease, mood disorder, and anxiety disorder.

(9.) Alternative models for count data with excess zeros include hurdle models and finite mixture models (Cameron and Trivedi 2005). Methods for modeling continuous cost data are reviewed by Mullally (2009) and Jones (2010).

(10.) Another reason for hospitalization is incurring injury due to accidents. A question on injury status is posed to all respondents in CCHS 2003 and 2005. but not in 2007, resulting in loss of a large number of observations. We therefore do not include an injury dummy in our reported regressions. As a robustness check we re-estimated the regressions with an injury dummy included using data for two years 2003 and 2005. The results are essentially the same and are .available upon request.

(11.) That occasional smoking is not associated with excess hospitalization is consistent with the possibility that such smoking may still cause premature death: if some occasional smokers die prematurely while in their seventies or eighties. rather than in their forties or fifties such effects will not be picked up in our analysis based upon the prime age group 25-65. We specifically focus upon the working years of the lifecycle given that our interest is upon discriminatory hiring.

(12.) The purpose of including obesity dummy is to emphasize that smoking is not the only costly observable health condition, and this latter now appears to be less costly than obesity. Note that our data have no measure of adiposity--an alternative metric of the impact on health of accumulated fat (Burkhauser and Cawley 2007).

(13.) We also estimated the same regressions using the larger sample of those aged 25-80. The results are very similar, and available from the authors upon request.

(14.) We also allowed the number of years of smoking to vary for current smokers (i.e., 0-10, 10-20, 20+ years) to see if the chronic conditions coefficients increase with the time length of their smoking. The results indicated that the variation in the chronic condition coefficients is really quite small, and where we do see variations, these variations are rarely outside of an overlapping confidence range for the coefficients.

(15.) This behavior need not characterize all recent quitters: sonic may decide to quit even if they are not suffering from poor health or under their physician's advice to quit. The regression coefficient captures that average impact of all individuals in the group.


* Neither author is in receipt of funding that would pose a conflict of interest. We are grateful to the editor and two referees for comments that substantially improved the paper.
COPYRIGHT 2014 Western Economic Association International
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2014 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Irvine, Ian; Nguyen, Hai V.
Publication:Contemporary Economic Policy
Article Type:Report
Geographic Code:1USA
Date:Oct 1, 2014
Previous Article:Welfare reform and children's early cognitive development.
Next Article:Fiscal convergence in the european union before the crisis.

Terms of use | Copyright © 2018 Farlex, Inc. | Feedback | For webmasters