Interwar Unemployment
Head Count

Interwar Britain remains one of the principal testing grounds for competing theories of macroeconomic fluctuations. To Keynesians, Britain's interwar experience provides incontrovertible evidence of pervasive involuntary unemployment due to insufficient aggregate demand. For nearly 50 years, interwar unemployment has been invoked as a classic illustration of the social costs of inappropriate fiscal and monetary management.

Until recently, there was little dissent from the Keynesian interpretation of interwar unemployment in Britain. In particular, it seemed impossible to reconcile New Classical explanations based on intertemporal substitution, according to which unemployment temporarily exceeds its natural rate when real wages are unexpectedly high, with recorded unemployment rates that between 1921 and 1938 scarcely dipped below 10%. In a series of recent articles, however, Benjamin and Kochin have argued that these sustained high levels of unemployment can be explained on equilibrium grounds once the effects of Britain's exceptionally generous unemployment insurance system are taken into account. Large numbers of workers opted for unemployment, Benjamin and Kochin argue, because the generosity of the dole reduced the opportunity cost of idleness.

Most responses to Benjamin and Kochin's analysis, both sympathetic and critical, have been based on regression analyses of aggregate annual data. These studies tend to find a positive partial correlation between the level of unemployment and the replacement rate (the ratio of benefits to average wages) as well as a regression coefficient which differs significantly from zero, using conventional levels of statistical significance. On this basis Benjamin and Kochin reject the null hypothesis that the unemployment insurance system was unrelated to unemployment during the interwar period. Critics of Benjamin and Kochin express reservations about both the data used, which may suffer from measurement error, and the interpretation of the regression coefficients, which may reflect causation running from unemployment to the replacement rate rather than vice versa and which may be sensitive to choice of sample period.

Many observers of this controversy will have concluded that empirical studies of interwar unemployment based on aggregate time series have reached the point of diminishing returns. These aggregate series are many steps removed from the individual and household level, where decisions were made and the effects of the Great Depression were felt. Progress in identifying the statistical significance and economic importance of the unemployment insurance system in the interwar period ultimately requires more and better data, ideally at the household level.

In this paper I report the results of research along these lines. The analysis makes use of the "New Survey of London Life and Labour", a survey of working-class London conducted principally in the years 1929-31 at the London School of Economics. I analyse the relationship between insurance benefits and unemployment status using a sub-sample drawn from some 27,000 case record cards completed in the course of New Survey. In this paper attention is restricted to adult males.

The results will not satisfy proponents of either the Keynesian or the New Classical view. The analysis of this paper indicates that interwar unemployment was neither largely a voluntary phenomenon nor entirely involuntary. After controlling for other characteristics of individuals, I find a generally positive if small association between unemployment incidence and the replacement rate. Although for adult males as a whole one cannot in all cases reject the null hypothesis of no association, one can use the standard errors of the coefficients to construct upper- and lower-bound estimates of the contribution to unemployment of the increasingly generous insurance benefits of the interwar period. This contribution is, however, small in the context of the Great Depression.

More revealingly, when one distinguishes between household heads and other adult males, one can identify a different relationship between the probability of unemployment and the level of benefits. Among household heads, for whom the responsiveness of labour supply should be relatively low, the effect of a change in the replacement rate on the probability of unemployment is essentially zero in the sample studied here. In contrast, among other adult males, the impact of a change in the replacement rate on the probability of unemployment is, in half the equations estimated, both statistically significant and economically important.

It would appear that secondary workers were more responsive than household heads to the incentives provided by increasingly generous unemployment benefits. It is plausible that the opportunity cost of idleness was lower on the margin for such secondary workers, whose families did not depend so heavily on their individual paychecks. In other words, they could afford to be responsive to the incentives to engage in additional search provided by an increasingly generous level of unemployment benefits.

But since such individuals comprise a relatively small share of adult males in the sample, the overall impact of benefits on unemployment tends to be small. I construct "counterfactual" unemployment rates on the assumption that replacement rates had been held at 1913 levels. The best estimates are those derived using the point estimates of the coefficient on the replacement rate. Calculations based on these point estimates suggest that the unemployment rate among household heads is changed little if at all but that the unemployment rate for other adult males can be quite considerably reduced. Since household heads comprise 77% of the sample of adult males, the overall adult male unemployment rate falls only moderately, from 8.6 to between 6.8 and 7.4%. While this change is non-negligible, its magnitude and its concentration among secondary workers convey a different impression from Benjamin and Kochin's results. By this measure, no more than a fifth of interwar unemployment, rather than Benjamin and Kochin's third to a half, is attributable to the generosity of the insurance scheme. Household heads, upon whose earnings so many wives and children depended for subsistence, show little responsiveness to variations in the replacement rate.

These point estimates alone, however, convey a false sense of precision. One of the advantages of basing calculations on regression coefficients is that they facilitate sensitivity analysis. Increasing and decreasing these coefficients by two standard errors yields lower- and upper-bound estimates which demonstrate the extreme sensitivity of the calculations to plausible changes in parameters. It is appropriate, therefore, to treat the evidence here on the relationship of benefits to the unemployment rate with considerable caution, but important to note also that this exercise in sensitivity analysis does not overturn the conclusion that the largest changes in unemployment rates are those experienced by non-household heads.

It would appear that Benjamin and Kochin's army of the voluntarily unemployed was more accurately a smaller and more mobile force composed of secondary workers.