Consumer’s Surplus with a Racial Apology? Black Relative to Non-Black Inequality in the Welfare Gains of Fuel-Efficient Cars and Trucks

The difference principle of Rawls (Rawls, p. 303; 1971) establishes as a principle of egalitarianism in welfare outcomes the idea that social and economic inequalities are to be arranged so that they are to be of the greatest benefit to the least-advantaged members of society.

Based on a sample of 1,279 households in the Consumer Expenditure Survey who purchased cars between 1983 and 1987, Goldberg (1996) finds no racial differences in the price paid for cars. She does however find that in the case of the price paid for vans and utility vehicles, relative to racial non-minorities, racial minorities received significantly lower discounts.

See for example Erdem, Senturk, and Simsek (2010), Diamond (2009), Kishi and Satoh (2005); Parry et al. (2007); Potoglou and Kanaroglou (2007).

This is the standard welfare-theoretic notion of willingness-to-pay (Cropper and Oates 1992; Spash, and Hanley 1995). Formally, one can view \( {p}_m^{c_i} \) as the minimum expenditure necessary to achieve pre-status quo welfare and consume one unit of q ^c without the market, and p _o ^c as the minimum expenditure necessary to achieve the pre-status quo welfare and consume one unit of q ^c in the market.

Of course, it is possible that an individual’s stated hypothetical preferences understates how he would behave if he actually had to make the choice/decision (Bohm 1972). This so-called hypothetical bias leads to stated preferences having an upward bias. However, there is evidence that this bias is not large. Murphy et al. (2005) found that in a meta-analysis of 28 stated preference studies that at the median, hypothetical values exceeded to actual value by approximately 35 %. As we are not trying to value cars/trucks specifically, if this upward bias is uniform across individuals in a survey, inferences about valuations conditional upon group membership (e.g. black, non-black), are still possible.

GSS data are available at http://​www.​norc.​org/​GSS+Website/​

Our parameterization of the τ _i follows that of Long and Freese (2001).

For a Multinomial Logit specification Pr(p = m | X) = F(τ _m  − Xβ) − F(τ _m − 1 − Xβ) = exp(Xβ _m|b ) − ∫  _j = 1 ^J  exp(Xβ _j|b )dX, where b is some base reference category.

Correcting for population size-related heteroskedasticity is one primary justification for using sample weights (Solon et al. 2013). In our parameter estimates, we allow for clustering on the standard errors across individual sensibilities and perspectives on environmental issues which varies in the population. This permits robust heteroskedasticity-consistent standard errors, allowing identification in the presence of non-equal probabilities of selection and the heteroskedasticity the results from sample selection.

Of course, an alternative is the Multinomial Probit model (MNP). Our choice of the MNL relative to the MNP is motivated by that fact the relative to the MNL, the MNP has likelihood function that tends to be flat near its optimum (Dow and Endersby 2004), which may yield weakly identified and arbitrary parameter estimates (Keane 1992; Alvarez and Nagler 1998.)

As in Cunha et al. (2007), in their consideration of Prescott and Visscher’s (1977) model of consumer choice, suppose in our case that individuals optimize a separable utility function U(A, Ο) = Aτ _i  + Ο, where A is a vector of cars/trucks, O is a vector of other goods, and τ is the consumer’s valuation of car/trucks. Let A _g + 1 be a fuel efficient car/truck, and let A _g be a standard non-fuel efficient car/truck. Individual i is indifferent between a fuel efficient and non-fuel efficient car/truck if A _g + 1 τ _i  − P _g + 1 = A _g τ _i  − P _g . This implies that in equilibrium τ _i  = c _g  = (P _g + 1 − P _g )/(A _g + 1 − A _g ). In this context, the threshold can be interpreted as the marginal price per unit. In this case, a fuel efficient car/truck is purchased if c _g  < τ _i  ≤ c _g + 1 If there are not enough individuals at the margins of the thresholds, and/or the thresholds are a function of unobserved individual characteristics, the demand function—which is a function of τ _i —will not be identifiable.

If indeed IIA is violated, the inclusion of choice-specific intercepts in a Multinomial Logit specification—which we include–will permit the estimation of choice probabilities (Train 1986; Dow and Endersby 2004).

The covariate summary in Table 1 reveals that approximately 1% of all respondents reported in the Strongly Oppose category. In the sample, no black respondents reported in this category.

The columns represent clustering on the variables representing knowledge of the causes/consequences of global warming as follows: (2) ARTSEALS, (3) INUITWAY, (4) NATENVIY, (5) NATMASS, (6) NOICECAP, (7) KNWGW, and (8) SEALEVEL. Column (9) weights the BL specification with the GSS weighting variable WTSSALL. This GSS weight takes into account the probability of non-response and the number of adults in a household—which measures the probability of response for a given household. In general, the GSS is self-weighting for household level variables, and we provide a weighted estimate so as to determine if the absence of blacks in the Strongly Oppose category affects parameter estimates.

Cohen (1988) suggested considering an effect size large if d≥.80 (approximately). For blacks and non-blacks, the mean value of GASREGS was approximately 1.43 and 1.20 respectively.

Our finding that CAFE standards are possibly not welfare optimal is in contrast to the findings of Goldberg (1998), which suggested that CAFE standards may be welfare optimal.

Subsidizing the purchase and use of motor vehicles is in fact a current and long-standing tax policy. Delucchi and Murphy (2008) estimate that the annual amount of this subsidy in the U.S. is upward to 64 billion dollars. Our results suggest that in principle subsidies can be tailored to equalize consumer’s surplus between blacks and non-blacks.