Attitudes Toward Public Policy Alternatives to Reduce Air Pollution

David A. Aaker, University of California, Berkeley
Richard P. Bagozzi, Massachusetts Institute of Technology
[ to cite ]:
David A. Aaker and Richard P. Bagozzi (1981) ,"Attitudes Toward Public Policy Alternatives to Reduce Air Pollution", in NA - Advances in Consumer Research Volume 08, eds. Kent B. Monroe, Ann Abor, MI : Association for Consumer Research, Pages: 616-621.

Advances in Consumer Research Volume 8, 1981      Pages 616-621


David A. Aaker, University of California, Berkeley

Richard P. Bagozzi, Massachusetts Institute of Technology

[The authors are grateful to the Field Research Corporation which originally collected the data used in this study, the University of California State Date Program which provided access to it, and the Institute of Business & Economic Research at the University of California which supported the study: Of course, these organizations are not responsible for the analysis or interpretation of data appearing herein.]

Research into public concern with environmental quality has mainly attempted to identify causes or determinants of environmental concern. Although there have been exceptions (Koenig 1975, Kinnear, et. al. 1974), the evidence suggests that concern is associated with higher levels of education and socioeconomic status, younger age groups, urban dwellers and nonminorities (see, e.g., Erskine 1972, Tognacci, et. al. 1972). With respect to political orientation, the evidence indicates that Democrats, liberals and those less politically alienated are more concerned with the environment (see, e.g., Tognacci, et. al 1972, Dunlap 1975, Koenig 1975, Buttell and Flinn 1978). There is also evidence that concern is related to perceived consumer effectiveness in influencing environmental quality (Kinnear, et. al. 1974).

This article will attempt to extend this research stream in several directions. First, the study will focus upon a specific domain of concern: concern for air pollution. Past research has tended to use a much broader definition of concern which would generate constructs like concern for the overall environment, ecology, or pollution. When the constructs consist of multi-item scales, the individual items are often quite diverse (e.g., Weigel and Weigel 1978). The problem with the use of such a general concern construct is that people can hold a variety of attitudes and beliefs toward the many components of the environment. If these components are combined, especially when they are combined in an unspecified way, the meaning of, and attitudes toward, the resulting construct may be unstable. Second, and perhaps most important, the research will study not only the antecedents of concern but also the impact of concern and other constructs upon the advocacy of a set of public policy alternatives such as closing freeways or adding additional pollution control equipment on automobiles. Previous research with few exceptions has been content to consider generalized concern as the sole dependent variable of interest. However, concern is more properly treated as an intervening variable. Inquiry should be extended to the link between concern and various public policy alternatives. The practical question is whether concern becomes translated into support for public policy alternatives that will be costly, inconvenient, or both. It is this link that is really important to understand the process by which an issue such as pollution gets addressed in our system of government.

Among the few studies that have looked beyond concern in order to investigate possible action that might be taken are several which use a "willingness to pay" construct (cf. National Wildlife Federation 1969 and 1970, Schuller and Ervin 1972). Although such a construct is somewhat artificial and therefore less useful than more realistic public policy alternatives, the results of these studies are suggestive. Basically, they indicate that, although a majority of people are willing to pay something to ameliorate problems like pollution, the amounts are likely to be quite modest. Thus, we might hypothesize that concern for problems of pollution may not necessarily be transmitted into support for public policy programs if those programs have associated costs. More particularly, the link between concern and support for public policy alternatives should be related to the costs of those alternatives for the respondents. A natural alternate hypothesis is that concern will indeed generate support for public policy alternatives.

Third, this study will introduce a variety of constructs which normally have not been considered in past investigations but which will act as important control variables, add useful exogenous variables, and introduce intervening variables enriching the model building process. The determination of the link between concern and support for public policy alternatives is potentially confounded by many factors which can be controlled. Thus, in this study, a liberal-conservative construct is included to control for individuals' inclinations toward governmental solutions; and a miles driven construct is included to control for each respondent's attachment and reliance upon the automobile. Useful exogenous variables included are objective measures of actual pollution levels. The few studies that have used pollution variables have found that these can influence attitudes (Degroot et. al. 1966). Several endogenous variables have the potential to increase our understanding of the attitude formation process. One such variable is the attitude toward the efforts of automobile companies to improve air pollution and another is beliefs as to the change of air pollution. Instead of the conventional two-way analysis, these variables are explored in the context of a structural equation model.

In sum, this study hypothesizes that whether people will advocate certain policy recommendations affecting themselves, industry, and government will depend upon (1) their beliefs that the problem in their immediate environment is getting worse, (2) their concern as to the seriousness of the perceived problem, (3) their political orientation, (4) their demographic characteristics, (5) their use of the automobile, and (6) the amount of actual pollution in their region. The particular sequence of relationships among these variables is developed below.


In May of 1973 (before the OPEC oil embargo), a total of 521 respondents were interviewed in their homes as part of the regular California Poll conducted by the Field Research Corporation. A cluster sampling design was used with over 100 randomly selected starting points for each cluster. Interviewing was conducted late afternoons and evenings on weekdays and all day on the weekend. Up to four call-backs were employed. One adult per household was selected systematically to provide a representative age and sex distribution. The sample was divided on a random basis into an analysis sample (n = 348) and a validation sample (nv = 173).

Exhibit 1 describes the eight endogenous variables and the eight exogenous variables developed from questions in the survey. The first five variables can perhaps be considered behavioral predispositions or policy variables in that they reflect respondents' positions on public policy legislative alternatives to combat air pollution. These include the installation of anti-pollution devices, gasoline rationing, restricting freeway use, encouraging rapid transit, and reducing automobile air pollution standards. Three alternatives were among those being seriously considered by state and local governmental units throughout the country who were under a mandate from the Clear Air Act of 1970 to submit plans for meeting the Environmental Protection Agency's air quality standards.



The CONCERN variable, a central variable in previous research, is based upon a question asking how serious is the air pollution problem. The BELIEF variable indicates the degree to which the respondent thinks that pollution has become worse during the past year. Finally, the AUTCO variable reflects the respondent's judgment as to whether the automobile companies have done all that they could to reduce air pollution.

The first exogenous variable represents the miles driven by the respondent, and the second reflects a liberal/ conservative self disclosure. It was felt that these variables should explain people's position with respect to public policy issues. Someone who drives more than average might feel that legislation is unnecessary or inconvenient. Further, a liberal orientation should lead toward a tendency to look to the government for solutions. These variables are not only interesting in themselves, but serve as important control variables when trying to determine the predictive ability of demographic variables.

The next three variables reflect the main types of air pollution in California. One interest is to see if one seems to be more dominant in the analysis than the other. The final three variables are the demographic variables of income, age, and education. The highest correlation among the exogenous variables was .34 (excluding the correlations among the three types of air pollution measures).


Stepwise regression was run upon the analysis sample for each of the endogenous variables. A recursive structure was used. The BELIEF regression used only the exogenous variables. The CONCERN regression also included the BELIEF variable as an independent variable. In addition to the exogenous variables the AUTCO regression included the CONCERN and BELIEF variables and the remaining "behavioral" predisposition variables included the BELIEF, AUTCO and CONCERN variables as independent variables. The recursive structure is shown in Figure A.

The results of the stepwise regressions are shown on the top line of each row in Table 1. Each variable coefficient reported has a beta weight of close to or exceeding 0.10 and each was at least significant at the 0.10 level. Also shown are the mean values for each variable and the R2 values.

The resulting model obtained from the stepwise regression was fitted to the validation sample using ordinary least squares. The results are presented on the second line of each row in Table 1. The numbers in parentheses are the t-values of the validation run. The results are summarized in Figure A. Only paths are included that have a beta weight of at least 0.15 on one of the data sets and were "confirmed" by the other. Stepwise regression with a validation sample is appropriate here since the prior theory is strong enough to support the model structure but is not strong enough to provide a priori specification of the paths. In fact, a purpose of the research is to identify paths. The low correlations among the exogenous variables and the use of a validation sample reduce the classic dangers of stepwise regression.





The Means

The means on the first four variables are directly comparable. A 2.5 rating indicates a neutral position. Two of the proposals, DEVICE and TRANSIT, obtained a neutral score and the other two were regarded negatively. [The differences between the means of the first four variables are statistically significant except for the difference involving DEVICE and TRANSIT.] Actually, the TRANSIT variable included a bus system proposal with exclusive bus lanes which was not supported (2.2) and a rail rapid transit system which was supported (2.8). During this time some exclusive bus lanes were tried in Los Angeles and proved to be unpopular. About 40 percent of the probability sample was from the Los Angeles area. Thus, the anti-pollution device and the rail rapid transit were judged neutral or favorable and the rest were judged unfavorably. Clearly, attacks on freeways and gasoline rationing are not popular among Californians who are heavily dependent on the automobile for transportation.

With respect to the AIR variable, respondents felt that the automobile standards were a bit too lenient but a small majority felt that they would still be relaxed if car performance were affected. The CONCERN variable indicated that respondents exhibited attitudes between feeling that the pollution problem was very serious (2) and somewhat serious (3). They further felt that the problem was (a) worse that it had been in the past (BELIEF) and (b) by a wide margin that the automobile companies could have been doing more (AUTCO).

The Intervening Variables

BELIEF.  BELIEF is a cognitive construct indicating whether air pollution is perceived to be worsening. It is hypothesized to be influenced by the pollution variables. In fact, the effects of the pollution variables were actually negative, although they were too small to be included in the table. Two qualifications need to be mentioned. First, the pollution variables did not reflect the change in pollution, only the absolute magnitude. Further, with minor exceptions, the areas had about the same levels of pollution level. Thus, the pollution variables are confounded with the "Los Angeles area effect." This interpretation problem is probably inevitable when dealing with pollution on an area basis as opposed to a neighborhood basis.

Those believing that air pollution has worsened tended to be younger as expected, but with lower income, which was unexpected. Perhaps the income variable served as a surrogate for neighborhood pollution levels. In any case, the R2 values were relatively low.

CONCERN.  The CONCERN construct has been the focus of many of the reported studies and is a central intervening construct here. Note that the R2 value for both data sets is over 0.20, an impressive level considering that disaggregative data are involved and that the dependent variable consists of a four point scale. If the respondents had been grouped, the R2 would be much higher. Bass, et. al. (1968) provide a discussion why R2 values are low for disaggregated data, and Aaker (1972) provides an illustration of what effect grouping can have on R2.

The two most significant coefficients have the expected sign. The pollution variable is large and positive as expected. Again, it should be noted that this variable is somewhat confounded by the fact that to some extent it is an indicator variable for the Los Angeles area. However, it seems safe to assume that a causal link should and does exist between pollution levels and CONCERN.

The BELIEF variable also has a large beta weight in both samples. Thus, there is evidence supporting a causal link between a perception that pollution is increasing and CONCERN. It could be argued that BELIEF is simply another measure of CONCERN and therefore what is observed is that two indicators of the same construct are correlated. Such a conclusion is regarded as not persuasive for two reasons. First, BELIEF is a perceptual or cognitive construct, while CONCERN is affective in content. Further, an attempt was made to consider BELIEF and CONCERN as indicators of a latent or unobservable variable in a recursive structure similar to that of Figure A. The resulting model had an extremely poor fit to the data indicating that such an approach was simply not compatible with this data. For a discussion of the use and testing of unobservable variables in structural equation models, see Asker and Bagozzi (1979).

It should be noted that NO2 was only marginally superior to DUST and OZONE, so the evidence is very weak indeed that NO2 is a more relevant pollution variable. In fact, the three variables have intercorrelations of .70, .86, and .93.

INCOME had a positive coefficient, although it was not significant in the validation sample. Still, it is of interest that income was included in the regression equation when such control variables as LIBERAL, MILES, AGE, and ED were also available.

AUTCO.  The most significant explanatory variable in the AUTCO equation is CONCERN. The R2 value with only CONCERN was nearly 0.09. Note also that the link between CONCERN and AUTCO is higher than the link between CONCERN and any of the behavioral predisposition variables except AIR. Thus, it seems that a concern about air pollution indeed seems to result in a tendency to "blame" the automobile companies for not solving the problem. The AIR variable also relates to this judgment since it involves the legislated automobile exhaust standards which would impose an automobile company generated solution. The other behavioral predisposition variables, on the other hand, involve more of a sacrifice on the part of the motorists.

The ED variable is positive, as expected, and the INC variable is also positive, though its influence disappears in the validation sample. The relatively weak performance of the demographic variables may, in part, be due to the existence of the LIBERAL variable which was significantly positive. To some extent, demographic variables might be simple surrogates for a liberal/conservative orientation.

The Behavior Predisposition Variables

Air.  Among behavior predisposition variables, the AIR equation generated the highest R2 value: 0.20 for the analysis sample and 0.12 for the validation sample. Again, it should be noted that these levels of R2 values are relatively high considering disaggregative data are used, the fact that the relationships involved are nontrivial, and the findings of previous studies (cf., Sharma, Kivlin, and Fliegel 1975).

A tendency to advocate stricter automobile exhaust standards is explained in large part by CONCERN (concern about the air pollution problem) and by AUTCO (the belief that automobile companies should have been doing more about the problem.)

A third significant variable is a pollution variable, NO2. However, the pollution variable has the wrong sign! Note that pollution also appears with a negative sign in the DEVICE AND RATION equations. The explanation for this finding is that the pollution variable is acting as a surrogate for the Los Angeles area and that other characteristics of residents of the Los Angeles area dominate or override the influence of pollution itself. This is somewhat surprising in view of the fact that MILES (a control variable for miles driven) is also in the equation. Thus, the fact that Los Angeles area residents drive more is not the characteristic that is causing the negative coefficient of the pollution variable. Note also that the coefficient of the MILES variable is negative as expected.

The remaining two explanatory variables, LIBERAL and INC, are in the expected direction but are not confirmed in the validation sample. The failure of the demographic variables to have more explanatory value is somewhat surprising.

The Remaining Behavior Predisposition Variables

The equations for the four other "behavioral" variables had smaller R2 values on the order of 0.10, and these values decreased in the validation sample. In the case of the TRANSIT and RATION variables, the drop in R2 was substantial. Still there are some worthwhile observations that emerge even for these two equations.

Overall the most significant explanatory variable is CONCERN. This finding is expected. In fact, it is somewhat surprising that the CONCERN variable did not have more explanatory power in the validation sample. Note that the BELIEF variable did not appear in any of the equations. This fact lends support to the hypothesized recursive structure. BELIEF seems to influence the behavior variables indirectly only through CONCERN and not directly.

The pollution variables appeared in the DEVICE and RATION equations (they were essentially zero in the others), but again with a negative coefficient. The most plausible interpretation is probably that the Los Angeles area residents do not support performance degrading automobile emission devices and gasoline rationing even after controlling for miles driven.

ED was consistently a substantial explanatory variable with the expected positive sign. Of interest is the fact that the other demographic variables failed to contribute. An exception is the large negative INC coefficient in the FREEWAY equation. The higher income people are opposed to radical efforts to cut down freeways whereas the lower income people are more disposed to such a move. This finding is unexpected. The hypothesis was that income would be associated with concern and with support of the policy alternatives. Thus, the direct negative links between income and other behavioral predisposition variables are somewhat surprising.

MILES and LIBERAL both made modest contributions in three of the equations. As expected, a willingness to support three proposals to restrict (RATION and FREEWAY) or affect (DEVICE) driving are negatively associated with miles driven. The support of the more radical government imposed solutions (RATION, FREEWAY and TRANSIT) is positively associated with LIBERAL.

Combining the Behavioral Predisposition Variables

It can be hypothesized that the five behavioral predisposition variables can be considered as indicators of a single underlying construct. Thus the model with a single unobservable "behavioral" construct was used with various combinations of the five behavioral predisposition variables as indicators. The results of this analysis (not shown) supported the counter hypothesis that the five variables are sufficiently different in the context of the structural model that they should not be considered indicators of a single unobservable. Any combination that included the AIR variable had exceptional poor fits as measured by a chi square value that is a by-product of the maximum likelihood estimation that was used (Aaker and Bagozzi 1979). When AIR was excluded, the fit was still unsatisfactory and, in fact, would improve dramatically each time the number of indicators was reduced. Only when one indicator was used was the fit extremely good. Hence, a separate recursive model for each of the behavioral predisposition variables was judged to be the most realistic model of the situation.


The results lead to the following summary comments and conclusions. First, education was found to have a direct although modest impact upon the policy variables. Income had a modest indirect effect through the CONCERN variable. However, the impact of these variables was perhaps smaller than expected, and the reason can be traced probably to the inclusion of the various control and intervening variables. Of note is the lack of impact of the AGE variable.

Miles driven had a direct impact on the automobile-associated public policy alternatives. The liberal/ conservative orientation variable had a direct impact upon the more radical alternatives involving substantial governmental actions.

The pollution variable did seem to have an impact upon the CONCERN variable. However, its direct impact upon the variables was negative due to other characteristics of Los Angeles area residents for which the pollution variables was acting as a proxy.

The hypothesis that the extent to which concern about air pollution generates support for public policy alternatives depends upon the cost of those alternatives is supported in this research. Concern about air pollution had a greater effect upon the AUTCO construct than upon the behavioral predisposition variables. This indicates that people were more ready to blame the automobile companies than they were to face certain alternatives involving sacrifices. Among the policy alternatives, the one with the strongest link to CONCERN was the proposal that the automobile air pollution standards should be stricter, a proposal that deflects the problem to the automobile companies and only indirectly affects the respondents. The DEVICE alternative, which also had a relatively strong link to CONCERN, was certainly much less extreme than the others. Of interest is the fact that TRANSIT, which was rated about as desirable as DEVICE, had a weaker link to CONCERN. Thus, the alternatives represented in the TRANSIT construct, although relatively palatable, were regarded positively in large part for reasons other then the fact that they represent a possible partial solution to the air pollution problem.


Aaker, D. A. (1972), "A Measure of Grand Acceptance," Journal of Marketing Research, 9, 160-67.

Aaker, D. A. and Bagozzi, R. P. (1979), "Unobservable Variables in Structural Equation Models," Journal of Marketing Research, 16, 147-158.

Bass, F. M., Tigert, D. J., and Lonsdale, R. T. (1968), "Market Segmentation: Group Versus Individual Behavior," Journal of Marketing Research, 5, 264-270.

Buttel, F. H. and Flinn, W. L. (1978), "The Politics of Environmental Concern: The Impacts of Party Identification and Political Ideology on Environmental Attitudes," Environment and Behavior, 10, 17-36.

Degroot, I., Loring, W., Rihm, A., Samuaels, S. W., and Winkelstein, W. (1966), "People and Air Pollution: A Study of Attitudes in Buffalo, N.Y.," Journal of Air Pollution Control Association, 16, 245-247.

Dunlap, R. E. (1975), "The Impact of Political Orientation on Environmental Attitudes and Actions," Environment and Behavior, 7, 428-454.

Erskine, H. (1972), "The Polls: Pollution and Its Cost," Public Opinion Quarterly, 36, 120-135.

Hummel, C.F., Levitt, L., and Loomis, R. J. (1978), "Perceptions of the Energy Crisis: Who is Blamed and How do Citizens React to Environmental-Lifestyle Trade-offs?" Environment and Behavior, `0, 37-88,

Kinnear, T. C., Taylor, M. R. and Ahmed, S. A. (1974), "Ecologically Concerned Consumers: Who are They?" Journal of Marketing, 38, 20-24.

Koenig, D. J. (1975), "Additional Research on Environmental Activism," Environment and Behavior, 7, 472-485.

National Wildlife Federation (1970), A Study of the Attitudes of the American Public toward Improvement of the Natural Environment, Washington, D.C.: The Federation.

National Wildlife Federation (1969), The U.S. Public Considers Its Environment, Washington, D.C.: The Federation.

Reizenstein, R. C., Hills, G. E., and Philpot, J. W. (1974), "Willingness to Pay for Control of Air Pollution: A Demographic Analysis," 1974 Combined AMA Proceedings, Chicago: American Marketing Association, 323-328.

Schuller, C. R. and Ervin, O. L. (1972), Public Attitudes and Regional Environmental Problems: Distributions and Correlates in a Southeastern City, Oak Ridge, TN: Oak Ridge National Laboratory.

Sharma, N.C., Kivlin, J. E., and Fliegel, F. C. (1975), "Environmental Pollution: Is There Enough Public Concern to Lead to Action?" Environment and Behavior, 7, 455-471.

Stalker, W.W. and Robison, C. B. (1967), "A Method for Using Air Pollution Measurements and Public Opinion to Establish Ambient Air Quality Standards," Journal of the Air Pollution Control Association, 17, t42-144.

Tognacci, L. N., Weigel, R. H., Wideen, M. F., and Vernon, D. T. A. (1972), "Environmental Quality: How Universal is Public Concern?" Environment and Behavior, 4, 73-86.

Weigel, R. H. and Weigel, J. (1978), "Environmental Concern: The Development of a Measure," Environment and Behavior, 10, 3-15.