Socioeconomic Distinction, Cultural Tastes, and Cigarette Smoking

Abstract

Objectives

The inverse relationship between socioeconomic status (SES) and smoking is typically seen in terms of the greater economic and social resources of advantaged groups, but it may also relate to cultural resources. This study aims to test theories of symbolic distinction by examining relationships between smoking and ostensibly unrelated cultural preferences.

Methods

Using the 1993 General Social Survey, ordinal logistic regression models, and a three-category dependent variable (never, former, and current smoker), the analysis estimates relationships of musical likes and dislikes with smoking while controlling for SES and social strain.

Results

Preferences for classical music are associated with lower smoking, while preferences for bluegrass, jazz, and heavy metal music are associated with higher smoking.

Conclusions

The results suggest that SES groups may use smoking, like other cultural tastes, to distinguish their lifestyles from those of others.

Socioeconomic Status and Cigarette Smoking

A well-known inverse relationship exists between education and cigarette smoking (Department of Health and Human Services (DHHS), 2001:35–37; Escobedo and Peddicord, 1996), a relationship that, given the harm to health from smoking, contributes to more general social disparities in health and longevity. Other components of socioeconomic status (SES) such as occupational prestige and income likewise have negative relationships with smoking (Barbeau, Krieger, and Soobader, 2004; Rogers, Nam, and Hummer, 1995), although not as strong as that for education (Winkleby et al., 1992).¹ For example, among persons 18 years of age and over in the United States in 2001, 27.5 percent of those with less than a high school degree smoked compared to 12.3 percent of those with a college education, and 31.4 percent of those below the poverty line smoked compared to 23.0 percent of those above the line (Woollery et al., 2003).

Identifying the explanatory mechanisms that lie behind the relationships (and thus to more general disparities in health and longevity) proves harder than describing the relationships. Lack of financial resources associated with low SES worsens health by limiting access to excellent medical care and safe living conditions, but the purchase of a product that harms health by those with fewer resources must involve something more. Perhaps better knowledge of the harm of smoking reduces the attraction to the habit among high SES persons. Yet, studies find little evidence that cognitive ability and understanding associated with education reduce smoking (Wray et al., 1998). To the contrary, knowledge of the harm of cigarette smoking is widespread (Viscusi, 1992), but higher educated persons respond more successfully to that knowledge (Barbeau, Krieger, and Soobader, 2004; DHHS, 2001:575).

Instead of increasing purchasing power or knowledge, high SES may bring other resources that provide both the motives and the means to avoid smoking. In terms of motives, high SES groups with the greatest earning power have the most to gain in longevity and lifetime wealth from a healthy, nonsmoking lifestyle (Hersch, 2000). Conversely, smoking may serve as a coping mechanism for low SES groups that face chronic strain because of their low social position and economic deprivation. If, as Wilkinson states (1996:186), “[s]moking has become a marker of socioeconomic stress,” then high SES groups that experience lower levels of such stress will gain fewer benefits from smoking than will low SES groups (Colby, Linsky, and Strauss, 1994; Lynch and Kaplan, 2000). Smoking may not only help cope with difficult economic and social circumstances, but also may have less serious health consequences for low SES groups. Those likely to die early from a variety of nonsmoking causes will see themselves as having less to lose from smoking and less reason to give up the pleasure of nicotine (Harris, Duncan, and Boisjoly, 2002; Lawlor et al., 2003). Given these incentives, advertisers target poor rather than affluent communities with billboards, promotions, sales outlets, and price cutting (Bassett, 2003).

In terms of means, education and, to a lesser extent, occupation and income increase resources, skills, abilities, and knowledge that help people confront and solve health-related problems (Mirowsky and Ross, 2003; Ross, 2000). Applied to smoking, these problem-solving skills may aid in resisting pressure from others to smoke, viewing attractive advertising images of smokers with skepticism, and overcoming the difficulties of ending an addictive habit (Bosma, Schrivers, and Mackenbach, 1999; Wray et al., 1998). In addition, SES increases social capital or the presence of cohesion, trust, reciprocity, and mutual aid in social relations (Kawachi and Berkman, 2000). Social capital provides both social support and social control (or positive and negative sanctions) needed to avoid or stop smoking.

Socioeconomic Distinction and Cultural Tastes

One other mechanism, despite having received little attention, might also prove important to both smoking and the relationship between SES and smoking. This mechanism emphasizes the significance of cigarette use as a symbol, fashion, or cultural taste that helps create boundaries between SES groups. Cigarette use or nonuse (and other health behaviors) may involve a conscious or unconscious effort of low and high SES groups to symbolically define a health lifestyle that distinguishes themselves from others (Cocker-ham, 2000). The key idea is that cultural sources of smoking and other health behaviors are correlated with but not reducible to economic resources.

Although rarely considered important for adults, cultural tastes relating to group membership, fashion, and distinction are critical to initiation of cigarette use among teens (DHHS, 1994). Teens often use smoking as a source of identity and image that both solidifies ties to friends who smoke and defines separation from others who do not smoke (Johnson and Hoffmann, 2000). Given common anti-smoking messages, teen peer groups in which smoking is common tend to hold more anti-authoritarian attitudes, have weaker ties to school and conventional institutions, and view themselves as different in image, goals, and interests (An et al., 1999; Jessor, Turbin, and Costa, 1998). Others, conversely, may use the rejection of smoking as a component of identity, peer-group norms, and more conventional behavior.

The same goals of following group norms and reaffirming group membership may help account for the observed inverse relationship between SES and smoking among adults. Rather than disappearing in adulthood, smoking-related norms and tastes may become more closely associated with SES distinctions. Such claims follow from more general and classical sociological arguments that have stressed the centrality of cultural distinction to social inequality. Near the turn of the century, Weber (1958) emphasized the use of lifestyles as a source of social prestige and a means of status groups to separate themselves from others. Highlighting the dynamic nature of cultural differentiation, Simmel ([1904] 1971) and Veblen ([1899] 1992) argued that upper classes first adopt innovative fashions and then move onto new fashions as lower classes imitate the upper classes. More recently, Bourdieu (1994) has maintained that distinctions based on cultural capital uphold class advantages, reproduce inequality, and promote domination and exclusion.

Cigarette smoking and nonsmoking may, in similar ways, represent forms of class distinction. As Griswold (1994:62) suggests, “[f]or baby boomers, the body—exercised, slimmed, well cared for—represented an ideal of youth and strength. High status was demonstrated not with martinis, fur coats, and silver cigarette cases, but with expensive mineral water, jogging, and disdain for smokers. As a cultural object for this group, the cigarette came to mean a foolish, and irritating to others, disregard for bodily health.” Consistent with the meaning attributed to the activity by high SES groups, smoking has increasingly come to be viewed by many as not only unhealthy but even immoral (Gusfield, 1993; Rozin, 1999). However, lower SES groups and those more marginalized from dominant groups would appear to differ in smoking norms. Based on current SES patterns of smoking, the habit may not carry the same stigma for lower SES groups as it does for higher SES groups—it might even signify risk taking, independence, and an anti-authoritarian attitude.

Some evidence suggests that cigarette smoking has followed the SES-based cycle of innovation, diffusion, and differentiation described by the classic theorists. Cigarette smoking, although viewed as unhealthy by many even before scientific evidence confirmed the belief, represented a technological innovation in terms of mass production and marketing (Griswold, 1994) and a social innovation in terms of the acceptability of public behavior (Ferrence, 1989). Earlier in the century when cigarette use was relatively new, higher SES groups, led ironically by physicians (Lopez, Collishaw, and Piha, 1994), first adopted the habit; use by lower SES groups came later. More recently, however, higher SES groups (again led by physicians) came to reject smoking, while lower SES groups continued to smoke at relatively high rates.² Such changes extend beyond cigarette smoking to include more general health lifestyles (Cockerham, 2000). Health behaviors such as exercise, moderate alcohol use, nutritious diet, appropriate body weight, and nonsmoking have become associated with SES.

These arguments about socioeconomic distinction, cultural tastes, and smoking have not been tested directly. Studies of the United States and Canada have demonstrated that the relationship between SES and smoking, once positive or near zero, has become increasingly negative over the past half-century (Escobedo and Peddicord, 1996; Ferrence, 1989; Pampel, 2004). Comparative studies across European nations (Pampel, 2002) find a stronger inverse relationship between SES and smoking for nations farther along in the process of cigarette diffusion than for those at earlier stages (where high SES groups more commonly smoke than low SES groups). Although offering insights consistent with the arguments, such data do not directly demonstrate a link between cultural tastes and smoking—a task made difficult by the fact that smokers and nonsmokers seldom consciously understand or admit their underlying motivations (Johnson and Hoffmann, 2000).

Testing the Cultural Argument

One way to test the cultural argument more directly comes from determining if well-known SES-based cultural tastes have a relationship to smoking (net of education, occupation, and income). If smoking represents a form of cultural distinction, it should overlap with other forms of cultural distinction. One particular cultural taste that varies with SES—musical likes and dislikes—offers a useful opportunity to examine possible overlap with smoking. Higher SES groups show attraction to high culture such as classical music and opera (Katz-Gerro, 1999; van Eijck, 2001) and to a wide variety of other genres that reflect their identity as cultural omnivores (Bryson, 1996; Peterson and Kern, 1996). Avant-garde music such as jazz and blues, despite its origins in the African-American community, has emerged since the 1950s as a high art form that also attracts highly educated devotees (Lopes, 2002). Lower SES groups show greater attraction to a fewer number of genres and more specifically to music with rural and working-class links, such as country and bluegrass. Youth-oriented and rebellious music such as heavy metal, rap, and reggae may likewise attract listeners from lower SES backgrounds.

Musical interests obviously do not cause smoking, but if both behaviors are components of SES-based lifestyles that status groups use as symbolic means of distinction, it implies an association (albeit spurious). The association should hold even with controls for SES, and indeed should reduce the negative effects of SES on smoking. For example, highbrow tastes for classical music and a diversity of genres should be associated with nonsmoking, as both represent cultural markers of high SES. In contrast, other musical tastes may have an affinity with the tendency today to use cigarettes as a symbol of rebellion, nonconformity, impulse and risk, and opposition to convention and comfort (Gusfield, 1993). Lowbrow tastes for country and bluegrass music should be associated with smoking because both reflect anti-elitist elements of rural and working-class lifestyles. Liking for anti-authority heavy metal, rap, and reggae music, by virtue of their distance from adult tastes and appeal to youth from lower-status backgrounds, should be associated with smoking, another behavior rejected by most adults and most common in lower SES groups. Liking for jazz and blues, however, contains contradictory tendencies. Although appealing to the more educated, jazz and blues music has maintained its avant-garde appeal and its distance from cultural orthodoxy (Lopes, 2002). This distance, combined with the performance of the music in clubs and bars rather than in concert halls, may make smoking more rather than less common among those liking jazz and blues music.

Cultural sources of differences in musical tastes and smoking involve more than SES. Musical tastes clearly differ by age, as might the relationship between musical tastes and smoking. Genres that appeal largely to youth such as heavy metal, reggae, and rap may prove important to smoking among younger age groups but do little to distinguish smoking among older age groups. Genres that appeal more to adults, such as classical or jazz, may show the opposite. If so, it suggests the need to examine relationships within as well as across age groups.

These arguments—that liking for classical music should be associated negatively with smoking, and that liking for bluegrass and country music, heavy metal, rap, and reggae, and jazz and blues should be associated positively with smoking—are admittedly ad hoc. The literature simply has not explored such relationships. However, the hypothesized link between smoking and ostensibly unrelated musical tastes (net of SES) provides a clearly falsifiable prediction that, if supported, would be consistent with a cultural explanation of SES differentials in smoking.

The predicted association between musical tastes and smoking in fact offers a stringent test of the cultural argument. First, critics have pointed out that (1) the correspondence between socioeconomic position and cultural tastes, although once important, has largely disappeared with the emergence of mass marketing of culture (Levine, 1988; Seabrook, 2001), (2) lower SES groups often generate innovations with regard to fashion, music, and art rather than imitate higher SES groups (Crane, 1999; Davis, 1992), and (3) cultural meanings are so fluid, complex, and individualistic that they are only loosely coupled to social and economic positions (Erickson, 1996; Lamont and Fournier, 1992). If, as these critics suggest, SES differences in musical tastes have become blurred, it will make it difficult to demonstrate a relationship with smoking.

Second, examining these hypotheses raises issues of timing. Ideally, cultural tastes should be measured when persons begin smoking, quit smoking, or decide not to begin or quit smoking. However, even longitudinal data present difficulties in matching smoking with its determinants because individuals face so many decision points. As a result, studies often use current characteristics of individuals to predict current smoking status, assuming that current characteristics reflect at least to some extent the characteristics at the time of smoking decisions. In support of this assumption, it seems likely that current musical tastes at least partly reflect similar interests at previous ages. To the extent they are stable, current cultural likes and dislikes should relate to earlier decisions to start, avoid, or quit smoking and therefore to current smoking patterns. To the extent that tastes change drastically, it will attenuate contemporaneous relationships and downwardly bias their association with smoking. If, despite this bias, support for the hypotheses still emerges, it will offer meaningful evidence.

Methods

Although a bit dated, the 1993 General Social Survey (GSS) contains measures of smoking, musical tastes, and sociodemographic position and offers a singular opportunity to test the hypotheses. Based on a stratified random sample of the noninstitutionalized adult U.S. population age 18 and over, the GSS relies on face-to-face interviews, a core set of questions answered by 1,606 respondents, and other sets of questions answered by randomly selected subsets of respondents (Davis and Smith, 1998). The smoking questions are asked of 1,057 respondents, and after deleting missing data, the sample size equals 1,046.

The surveys contain two smoking questions. One asks if the respondent currently smokes and another asks nonsmokers if they have ever smoked. The two questions allow creation of a three-category variable of never smokers (48.0 percent), former smokers (24.7 percent), and current smokers (27.3 percent). Validation studies find that self-reported cigarette use is generally accurate (Patrick et al., 1994). However, the GSS contains no information on age of initiation, age of cessation, or the number of cigarettes currently or formerly smoked per day. More detail on past or current smoking habits would allow more thorough tests of the hypotheses, but the categories of never, former, and current smokers capture commonly used and crucial characteristics of health-related behavior.

The GSS includes numerous background and SES measures relevant to cigarette smoking. In terms of background variables, sex takes the form of a dummy variable with men coded 1, and race takes the form of two dummy variables with blacks and others coded 1. Age measures decades from birth (1.8 to 8.9) and is treated as a quadratic to reflect the rise and decline of smoking over the lifecourse. Region of residence at age 16 takes the form of nine dummy variables created from 10 categories of similarly located states, and city size of residence at age 16 takes the form of a six-category variable ranging from country to large city. Because initial smoking emerges early in life, retrospective measures of region and city size of residence at age 16 more strongly influence smoking than current measures.

In terms of SES variables, education equals the respondent’s completed years of schooling. The relationship is nonlinear, such that the negative effect on smoking increases most at higher levels of education, and is best captured by a squared term alone rather than a linear term. Prestige of current or former occupation is coded on the basis of a scale constructed from ratings of the general social standing of occupations (Davis and Smith, 1998). Those with no information on current or former occupation (4.6 percent of the sample) are assigned the mean. Current family income in dollars comes from a 21-category measure ranging from under $1,000 to over $75,000, with the values recoded to the midpoint of the category. For the 9.5 percent of the sample lacking income data, the mean value is assigned. Mean substitution for occupational prestige and income allows use of complete data for the other variables and changes the estimated effects of occupation and income only slightly.³ All the SES components are transformed into standardized units to allow for direct comparison of coefficients.

A scale of social and economic strain extends the usual SES measures. It includes six items: financial satisfaction (satisfied, partly satisfied, not at all satisfied), change in financial situation (better, same, worse), unemployment in the last 10 years (no, yes), unemployment over the last five years (no, yes but not main earner, yes and main earner), number of traumatic events (deaths, divorces, unemployment, hospitalizations-disabilities) in the last five years (zero to four), and number of traumatic events in the last year (zero to three). The standardized scale constructed from the items has an alpha reliability of 0.734.

The cultural tastes variables come from a special set of questions on likes and dislikes for 18 musical genres that take values of 1—strongly dislikes, 2—dislikes, 3—mixed feelings or unfamiliar, 4—likes, and 5—strongly likes.⁴ Exploratory factor analysis with principal factor extraction and varimax rotation identifies four core sets of musical tastes. The factors correspond closely to those identified by Katz-Gerro (1999), but exclude one of the 18 genres, oldies, because it lacks clear meaning in terms of musical styles. Since oldies can connote 1960s rock, 1950s jazz, classic country, or big band music, liking for this genre overlaps with liking for nearly all the other genres. Including oldies in the factor analysis defines a fifth factor that has no relationship to smoking but, more problematically, weakens the distinctness of the other factors and their influence on smoking.

The results of the factor analysis in Table 1 define the four factors. The first represents highbrow musical tastes. It includes liking for classical music and opera—genres that require learning and discernment to appreciate. It also includes a wide variety of less popular music—musicals, big band, folk, and Latin—that reflect cultural openness of elite groups (Bryson, 1996; Peterson and Kern, 1996).⁵ The second factor includes musical genres often associated with youth and rebelliousness—heavy metal, rap, and reggae. New-age and contemporary rock have wider appeal but best fit within this category. The third factor includes bluegrass and country, genres that emerged from a rural, working-class tradition. The fourth factor includes jazz and blues. In the 1950s, jazz in particular developed into a high art form but, unlike classical music and opera, maintained its avant-garde character and opposition to cultural orthodoxy (Lopes, 2002).

TABLE 1.

Coefficients from Varimax Rotated Principal Factor Analysis^a

Musical Genre	Factor 1	Factor 2	Factor 3	Factor 4b
Classical	0.72	0.07	− 0.05	− 0.08
Opera	0.65	0.10	0.05	− 0.07
Musicals	0.64	− 0.07	0.07	− 0.25
Big band	0.52	− 0.16	0.14	− 0.27
Folk	0.47	0.03	0.40	− 0.01
Latin	0.46	0.21	0.07	− 0.28
Heavy metal	− 0.08	0.60	− 0.05	− 0.01
Rap	− 0.01	0.56	− 0.06	− 0.12
Reggae	0.17	0.53	− 0.05	− 0.30
New age	0.21	0.49	0.05	− 0.12
Contemporary rock	− 0.05	0.46	− 0.05	− 0.28
Bluegrass	0.17	0.00	0.59	− 0.08
Country	− 0.11	− 0.10	0.58	0.03
Jazz	0.28	0.21	− 0.05	− 0.61
Blues	0.20	0.16	0.15	− 0.58
Gospel	0.09	− 0.16	0.31	− 0.11
Easy listening	0.29	− 0.04	0.19	− 0.25

^aCoefficients >0.40 in bold.

^bScale created from factor is multiplied by −1 to indicate liking for jazz and blues.

Four standardized scales are constructed from the factor loadings. For simplicity, I refer to each factor in terms of the musical genres or audiences that load most highly: classical-omnivore, heavy metal-youth, bluegrass-country, and jazz-blues. The first factor should have a positive relationship and the other three should have negative relationships with smoking.⁶

To estimate the effects of these determinants on the three-category, ordered measure of smoking, the analysis uses ordinal logistic regression. A test of the parallel regression assumption required by the estimation technique (i.e., that the slopes, but not the intercepts, are parallel or equal across contrasts) compares effects of the variables in the model on never versus former and current smokers with the effects on never and former versus current smokers. The Brant test (Long and Freese, 2001:150–52) indicates that none of the slopes differs significantly at the 0.01 level, and only one (the dummy variable for other race) differs significantly at the 0.05 level. Given these results, the parsimony of an ordinal logistic regression model compared to a multinomial logistic regression model outweighs the minimal violation of the assumption. In the ordinal logistic regression, then, the slopes show the changes in the logged odds of being in higher relative to lower smoking categories due to a unit change in the independent variables. To transform the coefficients into a more meaningful metric, average changes in predicted probabilities for a unit change in the independent variables are presented with the logged odds.⁷

Results

The results first demonstrate that variation in musical tastes corresponds to social and demographic position. Table 2 displays the results of regressing the four musical tastes scales on the background and SES variables. The coefficients reveal that education has expected effects for three groups of musical genres: it raises the classical-omnivore scale, lowers the bluegrass-country scale, and raises the jazz-blues scale. It lacks influence only on the heavy metal-youth scale. Income raises liking for classical-omnivore music, social strain raises liking for bluegrass-country music, but job prestige has no net influence on tastes for any of the four musical genres. Also consistent with common expectations, classical-omnivore music proves most popular among females and older persons; heavy metal-youth music (including rap and reggae) proves most popular among young males, city residents, and minorities; bluegrass-country music proves most popular among whites, older persons, and smaller town residents; and jazz-blues music proves most popular among blacks, young persons, and city residents.

TABLE 2.

Unstandardized OLS Regression Coefficients and T Ratios (Absolute Values) in Parentheses for Models of Musical Tastes (N = 1,035)

Determinantsa	Classical-Omnivoreb	Heavy Metal-Youthb	Bluegrass-Countryb	Jazz-Bluesb
Sex: Male	− 0.31** (5.57)	0.15 ** (2.77)	0.02 (0.34)	− 0.06 (1.07)
Race: Black	0.01 (0.16)	0.24 * (2.56)	− 0.66** (6.76)	0.86** (8.76)
Race: Other	0.07 (0.55)	0.36 ** (2.69)	− 0.21 (1.43)	− 0.07 (0.52)
Age (decades)	0.18** (10.66)	− 0.24 ** (14.42)	0.09** (5.18)	− 0.05** (2.73)
City size at 16	0.02 (1.22)	0.04 * (2.28)	− 0.08** (3.82)	0.07** (3.26)
Educationc	0.33** (9.28)	0.06 (1.70)	− 0.12** (3.19)	0.13** (3.36)
Prestiged	0.02 (0.71)	− 0.06 (1.65)	− 0.02 (0.47)	− 0.02 (0.54)
Incomed	0.07* (2.30)	− 0.03 (0.87)	0.01 (0.40)	0.07 (1.93)
Social straind	− 0.01 (0.23)	0.05 (1.77)	0.08* (2.39)	0.05 (1.45)
Constant	− 0.30	0.71	− 0.14	− 0.21
R2	0.28	0.25	0.16	0.16

^*p<0.05;

^**p<0.01.

^aIncludes set of nine dummy variables representing region of residence at age 16.

^bStandardized factor scale from Table 1.

^cSquared and standardized.

^dStandardized.

The results next demonstrate that variation in smoking also corresponds to social and demographic position.⁸ The first equation in Table 3 includes the background variables (sex, race, age, city size, region⁹), the SES variables (education, occupational prestige, income), and social strain. Based on the coefficients from the ordinal logistic regression of the three-category smoking variable, education and occupational prestige reduce smoking, while social strain increases smoking. Social strain has the largest coefficient, followed by education, and occupational prestige. Without controlling for the closely-related measure of social strain, income also would have a negative effect on smoking of similar size to that of occupational prestige. Otherwise, males and those raised in cities more commonly smoke than females and those raised in rural areas. The age quadratic indicates a peak in smoking at age 47, which results from age patterns of quitting during adulthood and cohort differences in starting.

TABLE 3.

Ordinal Logistic Regression Coefficients and Z Ratios (Absolute Values) in Parentheses for Models of Cigarette Smoking

Determinantsa	Full Sample			Age-Specific Models
	Model 1	Model 2	Prob.b	Ages 18–39	Ages 40+	Differencee
Sex: Male	0.58** (4.62)	0.52** (4.09)	0.086	0.15 (0.53)	0.73** (4.80)	− 0.59 (1.88)
Race: Black	− 0.43* (2.03)	− 0.44* (1.97)	− 0.074	− 0.93 (1.93)	− 0.11 (0.42)	− 0.82 (1.48)
Race: Other	− 0.33 (1.03)	− 0.29 (0.90)	− 0.045	− 1.03 (1.58)	0.10 (0.25)	− 1.13 (1.48)
Age (decades)	1.10** (5.03)	1.08** (4.71)	0.175	0.89** 2.87	− 0.13* (2.17)	1.02** (3.23)
Age2	− 0.11** (5.12)	− 0.10** (4.74)	− 0.017
City size at 16	0.13** (3.08)	0.14** (3.11)	0.023	− 0.03 (0.35)	0.24** (4.32)	− 0.27* (2.52)
Educationc	− 0.29** (3.48)	− 0.23** (2.61)	− 0.038	− 0.50* (− 2.55)	− 0.18 (1.76)	− 0.32 (1.44)
Prestiged	− 0.19* (2.53)	− 0.19* (2.42)	− 0.031	− 0.20 (1.22)	− 0.20* (2.20)	0.00 (0.01)
Incomed	− 0.04 (0.49)	− 0.03 (0.43)	− 0.005	− 0.31 (1.66)	0.03 (0.33)	− 0.34 (1.65)
Social straind	0.39** (5.67)	0.38** (5.37)	0.062	0.56*** (3.84)	0.29** (3.47)	0.27 (1.62)
Musical Tastes
Classical-omnivored		− 0.17* (2.33)	− 0.028	− 0.16 (1.00)	− 0.16 (1.80)	− 0.01 (0.05)
Heavy metal-youthd		0.08 (1.05)	0.013	0.49** (3.30)	− 0.18* (2.05)	0.66** (3.88)
Bluegrass-countryd		0.18** (2.65)	0.030	0.26 (1.78)	0.22** (2.70)	0.04 (0.69)
Jazz-bluesd		0.15* (2.24)	0.026	0.09 (0.62)	0.21** (2.58)	− 0.12 (0.79)
Cut 1	3.27	3.17		2.17	0.42
Cut 2	4.47	4.40		2.81	1.91
Chi-square	155.38	173.94		90.27	123.29
Maximum likelihood pseudo-R2	0.14	0.16		0.26	0.15
N	1,035	1,035		298	737

^*p<0.05;

^**p<0.01.

^aIncludes set of nine dummy variables representing region of residence at age 16.

^bAverage change in predicted probabilities for unit change in each independent variable.

^cSquared and standardized.

^dStandardized.

^eCoefficient for ages 18–39 minus coefficient for ages 40+, and Z scores for test of equality of the coefficients in parentheses.

Three of the four musical taste variables, when added to the model in the second equation of Table 3, have the expected effects. Liking for classical-omnivore music lowers smoking, while liking for bluegrass-country music and liking for jazz-blues music raise smoking. The effect of the heavy metal-youth scale does not reach significance, however.

Transforming the effects into a more meaningful metric further demonstrates the influence of the musical taste variables. The third column of Table 3 lists the average changes in the predicted probabilities for a one-unit change in the independent variables (with the SES and musical tastes variables still being standardized). Social strain proves most important, but the musical tastes variables have impacts of 0.02 to 0.03 that are smaller than for education and similar to occupational prestige. For example, one standard deviation increases in education and liking for classical music have probability effects (at the mean of all the other independent variables) of 0.038 and 0.028, respectively.

Additional tests need to examine if relationships between musical tastes and smoking differ by age. The estimates in Table 3 for all ages combined might hide age-specific effects, particularly in the case of the heavy metal genre, which fails to reach statistical significance. To check for this possibility and to help capture the distinctiveness of youth, the next two columns of the table present estimates done separately for persons under age 35 and for persons age 35 and over. The last column lists the coefficient differences between the younger and older age groups and z scores for the significance of the differences between the coefficients (based on the standard error formula for coefficient differences presented in Brame et al., 1998:258).

Most coefficients do not differ significantly by age group. One exception is that liking for heavy metal and youth music has a significant positive influence on smoking among younger persons but not among older persons (where the effect reverses to become negative). This genre now also affects smoking but most clearly for younger persons (this same result appears when the sample is divided at age 30 or at age 40). The effects do not differ significantly across age for the other genres (although the effect of liking for jazz-blues is larger at older than younger ages). Thus, the positive effects of liking for heavy metal are limited to youth, while the effects for the other genres are best treated as similarly influential over all ages.

Discussion

Along with reaffirming the importance of SES for cigarette use, these results also identify an association with an ostensibly irrelevant factor—musical tastes. Liking for classical-omnivore music relates negatively, and liking for bluegrass-country, jazz-blues, and heavy metal-youth music relates positively to smoking. These variables are associated with SES but, also important, have independent influences on smoking. The effects depend on age, particularly in the case of heavy metal music, but overall musical tastes have associations with smoking that are slightly smaller than those of education and similar to those of occupation.

These results are consistent with claims that both smoking and musical tastes represent facets of class-based cultural norms. Musical preferences are spuriously (yet still meaningfully) associated with smoking through a possible common cause—cultural tastes that distinguish SES-based group membership. Although based on an inference about the underlying common cause, this finding is consistent with a cultural explanation of SES differences in smoking. Other resources prove equally or more important—SES variables and strain have strong effects on smoking. Yet, the independent influence of musical likes suggests the value of extending arguments about economic and social resources to include cultural tastes. The attention to cultural tastes is also consistent with the importance of identity and peer group for health behaviors, a point well accepted in explaining youth smoking.

This argument follows from theories of the importance of symbolic distinction to social inequality (Bourdieu, 1994; Simmel, [1904] 1971; Veblen, [1899] 1992). These theories view cultural resources as mechanisms of closure. Applied to health, SES represents a fundamental cause of disease that operates through multiple mechanisms and persists despite medical advances (Link and Phelan, 1995). Consistent with this insight, cultural resources as reflected in musical tastes appear to be associated with SES differences in smoking.

Alternatively, the relationship between musical tastes and smoking may reflect unmeasured traits that have little to do with socioeconomic distinction. Liking for classical music and avoidance of smoking may both stem from enhanced communication skills, knowledge, openness, and judgment gained from involvement in high culture (Mirowsky and Ross, 2003). The positive effect on smoking of liking for jazz-blues, a genre that attracts those with higher education, is less consistent with this possibility. Although the models control for education, the 1993 GSS does not have measures of personal control that can test the alternate argument directly. Still other unmeasured traits might be suggested to account for the observed relationships. While consistent with theoretical arguments about cultural tastes and socioeconomic distinction, the results allow for other interpretations.

The findings are further limited by the use of a 1993 cross-sectional survey to help understand a behavior that varies over time and across the lifecourse. Because musical tastes and smoking change with age, one-time measures of both concepts fail to tap attitudes and behaviors at younger ages when values and smoking habits are formed. The cross-sectional data also provide only a snapshot of the relentless societal transformation of cultural tastes. SES-based musical likes and dislikes have changed in the last decade—rap has crossed racial boundaries, some bluegrass styles have developed a high-art appeal, and smooth jazz styles have made their way to popular radio. Survey measures no doubt fail to fully capture the fluid nature of cultural meanings. In short, reliance on existing data limits the ability to match concepts with indicators, and the GSS measures of musical tastes and smoking have clear limitations. Despite these limitations, however, the effects of musical likes and dislikes on smoking prove surprisingly strong. Longitudinal data from more recent years and with more precisely tailored indicators of cultural tastes and smoking history would likely produce stronger rather than weaker results. If so, more attention is warranted for cultural components of health behaviors such as smoking.