The 1888–1890 Cost of Living Survey

The data in this article come from the “Cost of Living of Industrial Workers in the United States and Europe 1888–1890” survey (henceforth 1888CEX), conducted under Carroll Wright's direction when he was the U.S. Commissioner of Labor.⁵

The findings of the 1888CEX would inform the Congressional debate on the McKinley Tariff, and this is one of the reasons why the survey included households from both the United States and Western Europe in tariff-protected industries.⁶ The survey is broad in its scope, large in size, rich in detail, and by design allows for direct comparisons between American and European households. The 1888CEX data contain responses from 8,544 families working in iron, steel, coal, textile, and glass industries in both Western Europe and the United States.⁷ Nearly 80 percent of the sample, 6,809 households, was from the United States. The European subsample comes from Germany, Switzerland, Belgium, Great Britain, and France, although the vast majority, 1,024 households, came from Great Britain.

Contemporary observers have noted that, despite the ambiguity of the sampling procedure, the 1888CEX appears to be broadly representative of industrial households employed in the industries surveyed. Michael Haines notes that, in comparison to the age distributions of household heads for each industry in the Census of 1890 that “the broad similarities were more striking than the differences, and the case for the representativeness of the survey is strengthened [by the comparison to the 1890 Census].”⁸

John Modell further notes that the usual banes of household surveys, age-heaping and dollar-heaping, are not prevalent in the 1888CEX.⁹ He also finds that the household budgets are “careful documents, filled out with sufficient care to give one confidence in the outline they reveal.”¹⁰ Of the British sample, Lynn Lees holds that the British households in the 1888CEX are from “a large segment of the British labor force, one underrepresented in all other analyses of workers' standards of living in the late nineteenth century.”¹¹

The data contain detailed annual expenditure information for both food and nonfood items and annual income information for all members of the household (father, mother, and children). In addition, the data also contain demographic information on the household's age and sex composition, and detailed enumeration of the household head's occupation. Food expenditures are converted to quantities using the United States Senate Committee on Finance report “Retail Prices and Wages: A Report by Mr. Aldrich.”¹²

For each state I took the annual average price of each food item from June 1889 to May 1890.¹³ One of the benefits of the Aldrich report is that it lists more food items than the food categories in the 1888CEX. In these cases I assume that the expenditure on the food item listed in the 1888CEX was divided equally among the types of alternatives in the Aldrich report that could be classified under the 1888CEX listing.¹⁴ My estimates of the caloric elasticities, however, are not sensitive to the equal distribution assumption.¹⁵

I divide the expenditure on each food item by the price estimate for that food item in the household's state, which yields an estimate of food quantity. I then convert these quantities to calories with a calorie conversion table, taking into account any composite food items created in the quantity conversion.¹⁶

Summing these by food item I obtain the total annual calories available to the household. Although it is important to distinguish between calorie availability and consumption, all of the calculations are based upon total available calories—no wasting function is assumed.¹⁷ The Aldrich Report lists retail prices of the same food items for Great Britain, and similar conversions are used for the British sample.¹⁸ Specifically, prices are listed for Leeds, Manchester, and Liverpool in June of 1889.¹⁹ It is important to note that the price conversions for both samples come from the same report, which contains actual retail prices observed in stores, and the same composite foods were used where needed.

Throughout the article, per capita calories (PCC) are defined as the total annual quantity of calories available to a household divided by the household size.²⁰

Income and expenditure are also standardized, and per capita income (PCI) and per capita expenditure (PCE) are calculated by taking total income and expenditure and dividing them by the total number of people in the household, respectively. Boarders, a fairly common feature in American and British households, confuse the analysis. If boarders are included in household size the per capita income and per capita expenditure terms are understated, and therefore estimates of calorie elasticities are potentially overstated. To correct for this, I create “boarder corrected” per capita income and per capita expenditure variables by subtracting the number of boarders from household size used in the denominator of the per capita calculations. These corrected variables are noted as PCIB and PCEB, respectively, and provide lower-bound estimates of the calorie elasticity.²¹

Summary Measures and Methodology

Table 1 presents summary measures of the expenditure and caloric shares of the diet by food group for both the American and British samples. The most striking feature of the table is the apparent disconnect between expenditure shares and calorie shares. For example, expenditures on meats are a significant portion of total food expenditure (more than 25 percent of all food expenditures), and yet they comprise only 15 percent of the calories available to a household. Also, moving from the bottom to top deciles in terms of per capita income and per capita expenditure one finds the expected pattern of families moving away from cereals and towards meats. There is also a significant increase in both the expenditure and calorie shares of dairy as one moves from the bottom to top deciles. Although this result is not surprising, it does support Fogel's postulate that increasing income would lead consumers to choose calories that metabolized, thereby increasing their productive (work) capacity.²²

Table 1 also shows that American industrial households consumed roughly 250 more calories per person per day than did British households. Whereas American households have more than 1,650 calories per person per day, British households made do with roughly 1,400 calories per person per day. It is important to note, however, that mean per capita calories is an understatement of the calories available to these households because children are weighted equally to adults in the per capita calculation. After adjustment for the household composition I find that Americans consumed approximately 2,300 calories per adult male and the British consumed 1,800 calories per adult male equivalent. At baseline, then, there are significant differences in calorie availability between American and British industrial households. These differences also apply to the fraction of the budget devoted to food. Whereas 44 percent of the average American budget was devoted to food, more than 50 percent of the average British household budget was devoted to securing calories.

On the whole, Table 1 suggests that the diets of American and British families were fairly similar in terms of expenditure and calorie shares, although there were differences in the number of calories available to American and British households. At the summary level, these data are consistent with dietary studies by both economic historians and development economists. The patterns of expenditure and calorie shares are also entirely plausible given what we know about diets in these historical populations. Because these results are consistent both with the historical observations and with contemporary dietary studies, the case that these data are appropriate for caloric elasticity estimation is strengthened.

The income and expenditure elasticities of calories were estimated using ordinary least squares (OLS), assuming a constant elasticity function.

²³

The regressions take the form

where Z includes the size of the household, age and sex composition, geography, and industrial occupation of the household head. Note that because prices are held constant they collapse into the intercept. OLS regression estimates of the constant elasticity function are the primary object of interest in this article and are taken as summary measures of each population's demand elasticities for calories.

Comparing Calorie Elasticities

Elasticity values in this article are interpreted as the percentage increase in calories that would result from a 1 percent increase in income or expenditure. Given that calories are necessities, we should expect them to be income and expenditure inelastic. Intuitively, we would also surmise that a hungry person would devote a sizable portion of his increase in income to securing more calories. Similarly, we would generally say that households with high calorie elasticities are hungrier than households with lower calorie elasticities. In this way, calorie elasticities serve as an economic quantification of hunger.

From the theory of consumer demand, we expect to see a decline in the income elasticity of calories as income increases; eventually the demand for calories is saturated and the elasticity reaches zero. Also, we would expect the income elasticity of calories to be less than the income elasticity of food, because people can only consume a finite number of calories in any given period of time, although the price of those calories can vary considerably. It also suggests that the income elasticity of calories could approach zero for very high incomes, while for food it could remain greater than zero, although it would also tend towards zero at high-income levels.²⁴

This is another reason why the share of the budget devoted to food or average calorie intake may not be the best way to economically investigate hunger, and why the elasticities presented here enhance our understanding of nutritional well-being in the past.

Tables 2 and 3 report OLS estimates of the income and expenditure elasticities of calories for the American and British samples, respectively. Odd numbered columns provide estimates of the income elasticity of calories and even numbered columns provide estimates of the expenditure elasticity of calories. The specifications in columns 1 and 2 in the tables are the simplest estimates of the elasticities. The log of family size is added to the model in columns 3 and 4, and its inclusion suggests that the true elasticities are lower once family size is controlled for.²⁵

Columns 5 through 8 specify the fullest model, which controls for both family size and the age distribution of the household.²⁶

For each specification (columns 1–8 in Tables 2 and 3), I reject the hypothesis that the British elasticities are less than or equal to the American elasticities at all conventional significance levels.²⁷

One could argue, however, that such a rejection is not meaningful, and even misleading. As a glance at the regression results in Tables 2 and 3 will confirm, the elasticities are quite precisely estimated, and I am therefore bound to reject the hypothesis of equality even if the elasticities were economically (substantively) similar to one another. It is important, then, to establish the economic significance of the difference. Taking columns 1 and 2 of Tables 2 and 3 as instructive, the British income elasticity is more than 25 percent greater than the American value, and the British expenditure elasticity is more than 10 percent greater than the American value. All told, these differences suggest that British households would spend anywhere from 10 percent to 40 percent more out of marginal income on calories than American households for the same increase in expenditure or income, a difference that is economically large and significant.²⁸

These differences in nutritional well-being are further supported by other aggregate measures of well-being. According to Marianne Ward and John Devereux real per capita GDP in the United States was 21 percent greater than British per capita GDP in 1891.²⁹

For the five years 1887 to 1891 American nominal per capita GDP was, on average, 20 percent greater than British per capita GDP.³⁰ Data on height provides further evidence of the plausibility of these historical nutrition measures. The British birth cohort of 1862, who would have been in their mid-twenties at the time of the survey, were on average 169 centimeters tall whereas the American 1860 birth cohort was 172 centimeters tall.³¹ The 1890 American birth cohort was 169 centimeters tall, whereas in 1900 British stature averaged 167 centimeters.³²

Further nutrition-based evidence bolsters the plausibility of the calorie demand elasticity differential. D. J. Oddy notes that Drummond argued that British malnutrition at the time was “more rife in England than it had been since the great dearths of medieval and Tudor times.”³³

Oddy also notes that among British World War I recruits, who would have been born in the late nineteenth century, only 36 percent were above 167 centimeters in height and 130 pounds in weight, whereas Haines and Steckel found that the average American World War I recruit was 171 centimeters tall.³⁴ This evidence certainly suggests poorer living standards in Britain in the late nineteenth century, and the elasticities not only confirm this, but add another dimension to the story.

What is most interesting about these elasticity differentials is that they strengthen the general argument that, in a cross section, significant calorie elasticity differentials imply significant nutritional differentials between subsamples. The narrative description of American and British working-class nutrition strengthens this interpretation. Peter Shergold found that British families were nutritionally worse off than American families.³⁵

He explained that British workers “consumed less of almost every [food] article than his American counterpart. The major exception was bread,” with his analysis showing that British families consumed more tea and rice than American households as well.³⁶ For every other food item Shergold reports, however, the British were left lacking, and their calorie elasticities reflect their relatively malnourished state in the late nineteenth century.

Other investigators found similar patterns among industrial households in the late nineteenth century and early twentieth century when looking at detailed household budgets. Although both American and European laborers were malnourished, observers noted that European industrial workers were more malnourished than their American counterparts. For example, Frank Streightoff estimated that one-third of all industrial households in the United States at the time were malnourished and that only a small fraction (less than 10 percent) received an adequate amount of protein.³⁷

He conceded that “it may be that the American laborer has a larger variety of better food than his European brother, but this fact does not ameliorate his condition.”³⁸ This evidence points to a relatively higher standard of living for American industrial workers in the late nineteenth century. Furthermore, the evidence suggest that these differences in nutritional well-being are relative indeed—from all accounts both American and British industrial workers in the late nineteenth century were malnourished by modern standards. The press of hunger for the British industrial class was clearly greater than that of the American industrial class, insofar as calorie elasticities are a marker of hunger. When one considers Lees's description of the households in the British data these results may well understate that differential, as the British households may represent a higher segment of British industrial society than the American sample.³⁹

Relaxing the Parametric Assumptions

OLS estimates are only part of the story, however. Although the constant elasticity function has much to commend it, many have argued that the assumption of a constant elasticity is rather restrictive. Earlier, I posited that the calorie income and expenditure elasticities should decrease as income and expenditure increase, and yet the constant elasticity function, by assumption, negates this possibility. Nonparametric techniques are available to relax the constant elasticity function assumptions to make them more consistent with economic theory. Locally linear regressions were used to nonparametrically estimate the relationship between calories and income.

The locally weighted regression assumes that the functional relationship between y (calories) and x (income or expenditure) is linear within small local neighborhoods of x.⁴⁰

Locally weighted regressions in this article are weighted regressions where the local neighborhood is the 30 percent of the total observations that are closest to each individual x_j.⁴¹ Because a linear function within a small neighborhood will closely approximate nearly any well-behaved function if the neighborhood is sufficiently small, its slope will approximate the local elasticity. The procedure has the added advantage of not imposing a parametric variation on the elasticity that may or may not be true in the data. The locally linear regression leaves us with a function that need not be globally linear. To obtain local estimates of the elasticities I divided the x axis for each respective sample into 100 evenly spaced points along the x axis and calculated the slope between those points.⁴²

Figure 1 shows the locally weighted regression fitted values from the log of per capita calories on log of per capita income regression, and Figure 2 shows the fitted line for the local log of per capita calories on log of per capita expenditure regressions for both samples. As both Figures 1 and 2 show, the American and the British functions are nearly linear in both specifications, even with the relatively small neighborhoods in the local regressions, and this suggests that any potential information loss from the parameterization required by OLS will not be large. If we can take the slopes of these lines as rough estimates of the elasticities, we see that in Figure 1 the slope of the American line is close to 0.4, whereas for the British line it is close to 0.6, and in Figure 2 the slope is 0.6 for American households and 0.7 for British households. The steeper slopes of the British curves would imply that the British elasticities are greater than the American elasticities. Also note that the slope in Figures 1 and 2 are close to the elasticity estimates from the simple lnPCC on lnPCI, lnPCE regressions in columns 1 and 2 of Tables 2 and 3.

Figure 3 shows the estimates of the local income elasticities of calories for the American and British samples. A six-degree polynomial trend line is included on the figure as well, and this trend line confirms that the elasticity does indeed decline with income. For the British households, the elasticity goes from a high of 0.7 at low incomes to a low of 0.55 at high incomes. For American households the elasticities go from a high of 0.6 to a low of 0.2. Figure 4 shows a similar trend for the expenditure elasticity of calories, although the expenditure elasticities are greater than the income elasticities. Both the American and British households have expenditure elasticities near 0.85 at low expenditures. The low point of expenditure elasticities of 0.65 for British households is seen at moderate expenditures, whereas the low point of expenditure elasticities for American households of near 0.35 is seen at high expenditures. Note that in both Figures 3 and 4 the elasticities are centered around the slopes of the lines in Figures 1 and 2, and these midpoints agree with the elasticity estimates in column 1 of both Tables 2 and 3. Overall, the nonparametrics show that these elasticities appear to behave as economic theory would predict.

Income Composition, Industry, and Calorie Elasticity Differentials

We should be mindful of the fact that the OLS results reported above are aggregate estimates of the elasticities and act as a summary measure of the population's elasticity. Table 4 presents results of the calorie elasticities estimated by household income composition.⁴³

A glance at the table reveals an interesting trend. For British families, the wife or child working results in a lower elasticity estimate than is seen when only the husband is working, but for the American households the trend is the opposite. It is not surprising, then, that the hypothesis that the elasticities are equal is only rejected for the subsample where only the husband earned income outside of the household. Because this group is, by far, the largest we could say that the elasticity differentials are driven by the husband's income. This general trend seems to imply that American wages were such that when wives and children participated in the labor force it was to make up the income gap created by the husband's labor market outcome.

Summary measures of income by type of income shed more light on the issue than the regression results reported in Table 4. Americans were wealthier than their British counterparts, but the difference narrows for the subgroups where wives and children work outside of the home. Although the overall United States-Great Britain per capita annual income differential is $34.55, for households where the husband is the only wage earner the difference grows to $50.79. This differential decreases to $9.56, however, when I compare the subsamples where children are working outside of the home and becomes negative (–$13.76) when wives work outside of the household. There is a substantial loss of income relative to British households when American wives and children work outside of the home. These facts, when viewed in conjunction with the results of Tables 2 and 3, imply that the elasticity differential is driven by the relatively high earnings of American men.

It would be unwise to argue that demographic composition accounts for the calorie elasticity differential when income composition between the two samples differs. British families were much more likely to send their children to work than American families. Whereas less than 30 percent of the American households had children who worked outside of the home more than 40 percent of British families had children in the labor market. Also, American fathers earned less than British fathers when their children worked outside of the home, although the total income ratio between countries for this group remains nearly equal to that for all families in the survey (120.41 percent versus 128.8 percent for all families). Also note that household size and age composition are included in the fullest specifications in Tables 2 and 3, and the elasticity differential remains.

Historical observers noted this trend as well. Streightoff and Robert Chapin held that American families with multiple wage earners were worse off than their single-earning counterparts among industrial households at the time.⁴⁴

Chapin, in particular, states that families with multiple wage earners were more likely to be malnourished than one-income households. David Angus and Jeffrey Mirel found that children in Europe played a vastly different role in their households than American children did, and therefore the differences that I attribute to economic demography in fact reflect more on institutional differences.⁴⁵

Table 5 displays the industry comparisons between American and British households.⁴⁶

In all but two industries, pig iron and wool textiles, both the income and expenditure elasticities of American households are lower than those of British households in the same industry. The differentials do vary a great deal, but the broad pattern is consistent with what has been noted earlier. Also, in all but one industry (cotton textiles), per capita income and expenditure, total income and expenditure, and husband's income were greater for American households, although once again magnitudes vary.

Due to small sample sizes for each industry, hypothesis tests are not very instructive as they have relatively little power; if I fail to reject the hypothesis of elasticity equality it could still be true that the elasticities are not close to one another. Nevertheless, I bootstrapped the sample variances of the income and expenditure elasticities of calories by industry, which were calculated with an OLS regression of the log of per capita calories on the log of per capita income and the log of per capita expenditure, respectively, so such a test could be performed with the best possible estimate of the standard errors. The rejection of the hypothesis that the British elasticities are less than or equal to the American elasticities is fairly robust to the within-industry comparison, which tells us that the results presented here are not due to the industrial composition of the data.

There were, however, large differences in the size of the calorie elasticity differential by industry. In some industries, such as glass and steel, the hypothesis was rejected at all conventional significance levels. For other industries, such as cotton textiles, the hypothesis is rejected only at the 10 percent significance level for the income elasticity, and not rejected for the expenditure elasticity. The general trend, however, was one of rejection of the hypothesis at least at the 10 percent significance level when the point estimate of the American elasticity was less than the British point estimate of the elasticity. Because the tests lack power, however, it is not clear how a failure to reject the hypothesis that the British elasticities are less than or equal to the American elasticities should be interpreted.

In addition to calorie elasticities, Table 5 also reports the expenditure elasticity of food. The differences in the food elasticity, for the most part, mirror those of the calorie elasticities, but they are usually much greater than the calorie elasticities. There are some important distinctions, however. In the Bar Iron industry, for example, the food elasticities are close to one another but the calorie elasticities are relatively far apart. This may be a situation where inference based upon the food elasticity will be different from the conclusions drawn from the calorie elasticity. As in the case of household income composition, the hypothesis that the American and British caloric elasticities are equal to one another is rejected when the per capita incomes of the British and American subgroups were highly dissimilar.