Introduction and Summary:
In “The Role of Prices in Measuring the Poor’s Living Standards,” Christian Broda, Ephriam Leibtag, and David E. Weinstein (2009) use proprietary data — the 2005 Nielsen Homescan dataset — to analyze differences by income level in the prices paid for food. They find that Nielsen households with incomes above $60,000 pay somewhat more for the same food items than most households with lower incomes, with Nielsen households with incomes above $100,000 paying the most. Based on this finding and additional regression analyses, they conclude broadly that the “poor pay less — not more — for the goods they purchase” and that not accounting for this suggests that income inequality may be between 2.5 to 5 percent less than shown by national statistics.
This paper reviews the methodology and findings of Broda et al. and concludes that:
1) Research that relies on Nielsen data to provide information about the relative prices paid by low-income households in the population, including Broda et al.’s conclusion that “the poor pay less,” is unlikely to be reliable. Both low- and high-income households appear to be underrepresented in the weighted and unweighted versions of the Nielsen dataset. Moreover, both the low-income and high-income households in the dataset are almost certainly unrepresentative of their counterparts in the population as a whole. In fact, the methodology almost seems designed to select the subset of lower-income households who are careful comparison shoppers, and to exclude those lower-income households who are likely to have the greatest barriers to economizing on food, particularly those with disabilities, low literacy levels, and limited or no access to the internet. These differences could account entirely for Broda et al.’s main finding that lower-income households pay less than higher-income ones for food.
2) Even if we ignore the fact that Nielsen households are almost certainly unrepresentative of households in the population as a whole, Broda et al.’s interpretation of their results as showing that the “poor pay less” than higher-income households for food is overstated. In addition to the income-price differences noted above, their results also show that households in the two lowest-income categories in the Nielsen data set (households with incomes under $5,000 and those with incomes between $5,000 and $7,999) appear to pay modestly more for food than most other low- and moderate-income households with incomes just above them. Based on the results they present, we can’t say whether these two income categories in the Nielsen data represent a modest share of the “officially poor” or a more substantial one, but clearly some significant share of “the poor” pay more for food in their results. Moreover, a comparison of the income categories in the Nielsen dataset with those from other representative national surveys suggests that the bottom-two income categories in the Nielsen database likely represent a substantial share of officially poor households.
3) Broda et al.’s “poor pay less” interpretation of their results is also overstated because it doesn’t account for additional costs that are likely to be disproportionately borne by economizing lower-income households — these costs include time spent comparison shopping, time spent traveling to more distant stores, and inconveniences associated with buying in bulk (including storage costs and wastage) — or for the additional benefits related to store amenities that are almost certainly disproportionately realized by higher-income households. Thus, even if we again ignore the fact that Nielsen households are almost certainly unrepresentative of households in the population, we should be very cautious about concluding that the differences Broda et al. find in the prices paid per food item by household income level correspond to differences of a similar magnitude in the welfare or living standards associated with those items.
4) While Broda et al. conclude that higher food prices paid by higher-income households mean that income inequality is overstated in standard income statistics — that is, that income differences in food prices should be viewed as a factor that moderates the real level of inequality — it seems more likely that the massive increase in income inequality over the last several decades has directly contributed to income-based differences in food prices. Between 1979 and 2003, the real incomes of families in the top fifth of the income distribution grew by 49 percent. By comparison, the incomes of families in the bottom fifth increased by only one percent, and those in the middle quintile by only nine percent. As a result, low- and moderate-income households have likely felt greater pressure to economize in food and other areas than they would have if their incomes had grown at a similar rate as incomes did for higher-income households.
5) Whether or not the poor pay more for food purchased for consumption at home is to day a relatively small part of the broader and now decades-old question of whether “the poor pay more.” In 1960, food expenditures accounted for about one-quarter of the average family’s expenditures; today, they have dropped to nearly half that (about 13 percent). The food expenditure share of low-income households today (16.2 percent) is slightly higher than the average for all households, but the trend is the same. Researchers seeking to answer the question of whether the poor pay more today need to consider other factors, particularly transportation costs and financial services, that have taken on increased importance in recent decades.
6) Higher-income households likely benefit disproportionately from the now-widespread acceptance of credit cards by stores selling groceries. Grocery stores neither provide a discount for shoppers who pay cash nor require credit-card users to pay an additional amount to cover the costs to food retailers of accepting their credit cards (roughly $2 for every $100 of credit-card purchases). As a result, these costs are passed on in the form of higher prices for all food shoppers. To the extent that higher-income households are more likely to use credit cards for grocery purchases than lower-income households — which seems quite likely given general data on income differences in the use of credit cards — the generally poorer customers who don’t use credit cards (and do not receive the benefits associated with them) are subsidizing the purchases of the generally richer customers who do.
7) In a short section of their paper, Broda et al. argue that the official poverty rate overstates actual poverty today by some 60 percent because of biases in the Consumer Price Index. The current poverty line implied by this finding — roughly $11,025 for a family of two adults and two children, rather than the current $22,050 — strains credulity. There is no reason to believe that a family of four can live a minimally decent life in 2010 on $11,025 a year.
The following section of this paper summarizes Broda et al.’s methodology and findings. Section II examines the representativeness of the Nielsen dataset, especially in regard to low-income households. Section III discusses whether Broda et al‘s findings actually support their conclusion that the “poor pay less,” even if we ignore the unrepresentativeness of the Nielsen dataset. Sections IV and V raise some additional issues related to the relationship between inequality and food prices, the distribution of the costs and benefits of credit card use for food purchases and the extent to which the poor pay more for financial services. Finally, Section VI discusses the distinct issue, raised briefly by Broda et al., of whether the current poverty thresholds ($22,050 for a family of four) are more than double what they should be.
Shawn Fremstad is Director of the Inclusive and Sustainable Economy Initiative at the Center for Economic and Policy Research in Washington, D.C. This article was first published by CEPR in December 2010 under a Creative Commons license.