Unemployment Incidence in Interwar London

Transcription

1 Economica (2002) 69, Unemployment Incidence in Interwar London By TIMOTHY J. HATTON and ROY E. BAILEY University of Essex Final version received 1 August The causes of unemployment incidence in interwar Britain have been the subject of much debate since Benjamin and Kochin claimed that it was due largely to generous unemployment benefits. We use the records for 30,000 workers from the New Survey of London Life and Labour ( ) to estimate the determinants of unemployment incidence. We find no significant effects of the benefit wage ratio on the unemployment probability for adult males when we allow for skill and industry effects. Separate regressions for younger males and for females also fail to reveal significant effects from unemployment benefits on the pattern of unemployment incidence. INTRODUCTION Between 1920 and 1938 the unemployment rate in the United Kingdom averaged 14%, and the period has been viewed as one of macroeconomic turmoil and policy failure. Until the late 1970s the literature on the era focused on the demand side of the labour market and on government policy: the return to the gold standard in the 1920s and responses to the Great Depression of the 1930s. 1 An alternative view, more popular among historians than economists, is that unemployment in the interwar period was essentially structural. 2 High rates of unemployment in the traditional staple industries was, to some degree, reflected in regional unemployment rates with much higher rates in the north of England, Wales, Scotland and Northern Ireland than in the southern regions. 3 A combination of low rates of labour mobility and a lack of transferability of skills limited the degree of labour market adjustment throughout the interwar period. It was against this background that Benjamin and Kochin s (1979a) article created a wave of controversy. According to them, The army of the unemployed standing watch at the publication of the General Theory was largely a volunteer army (p. 474). This, they argued, was due to the high rates of benefit, relative to wages, provided by the unemployment insurance system, combined with the liberal eligibility conditions under which these benefits were administered. The main piece of evidence they offered was a time series equation for the years demonstrating a positive correlation between the benefit wage ratio and the unemployment rate. 4 Benjamin and Kochin viewed benefit-induced unemployment in the interwar period as the result of individual behaviour rather than of collective action. They argued that the effects of the unemployment insurance system could be clearly observed in the differences in unemployment incidence among individuals or groups who faced different benefit wage ratios. In particular, they claimed that juveniles (those aged 16 17) and young adults (aged 18 20) had lower unemployment rates because they faced lower benefit wage ratios. For women, lower unemployment rates could be explained both by lower benefit wage ratios and

2 632 ECONOMICA [NOVEMBER (especially for married women) by restricted access to benefit following a change in regulations in Hypotheses about the incidence of unemployment cannot be tested using aggregate time series data. Time series models typically use a representative benefit rate such as that for a man with a dependent wife and two children. In order to address the question of incidence across different groups, which was so central to the Benjamin Kochin argument, individual level data are indispensable. The approach using individual level data, now the standard form of analysis for more recent periods, has been adopted only in the pioneering study of Eichengreen (1987). 5 Not surprisingly, this study has been widely cited as the best available evidence on the effects of the interwar unemployment insurance system on individual incentives. Eichengreen took a 10% sample from the surviving records of the New Survey of London Life and Labour (NSLLL), a survey of working-class households in London undertaken in He estimated probit equations for unemployment incidence among males aged 18 64, dividing these into household heads and non-heads. Unemployment incidence was explained by the benefit wage ratio together with a number of individual and household characteristics. The results indicated that there was a positive relationship between unemployment incidence and the benefit wage ratio but that the effect was significant only for non-household heads. On the basis of these coefficients, Eichengreen estimated that the impact of the interwar unemployment insurance system was modest, raising the overall unemployment rate by about 1.2 percentage points. He concluded that Benjamin and Kochin s army of the voluntary unemployed was essentially a squadron of secondary workers (1987, p. 4). Although it improves on previous analyses, Eichengreen s study was limited by the constraints of his sample. Our recent computerization and coding of the entire set of records from the NSLLL makes available a larger set of data and a wider range of variables. Using this source, we are able to improve on Eichengreen s analysis in three ways. First, with a larger data set we are able to analyse unemployment incidence separately for women and for young workers. Second, we are able to control for the effects of occupation and skill level. Third, knowledge of the dates of the interviews allows us to calculate an improved measure of the unemployment benefit rate relevant to each individual in the survey. The remaining sections of the paper are organized as follows. First, we examine the structure and operation of the interwar unemployment insurance system. We then introduce our data source, the NSLLL, and discuss the variables derived from it. The sections that follow present our main results, first for adult males aged 25 64, then for younger males aged 16 24, and finally for females. The findings are summarized in the concluding section. I. THE INTERWAR UNEMPLOYMENT INSURANCE SYSTEM In order to assess the effects of the unemployment benefits on unemployment incidence, it is important to examine the characteristics of the unemployment insurance system. First established in a narrow range of industries in 1911, it was substantially widened in 1920, and by 1931 it covered about two-thirds of

3 2002] UNEMPLOYMENT INCIDENCE IN INTERWAR LONDON 633 the labour force, but only about half of all women employees. The excluded occupations included agriculture, forestry and fishing, domestic service, public service workers (such as the armed forces, police, teachers and established civil servants) and non-manual workers whose earnings normally exceeded 250 per year. Workers under the age of 16 or 65 and over were excluded from insurance. For covered workers who became unemployed, benefits were given at a flat rate, unrelated to the previous wage or occupation. Rates of benefit were graduated by age up to 21 with lower rates for women than men. In addition to the basic rate, standard allowances were given for dependants such as spouses who were not working and were not insured in their own right and dependent children. During the period covered by our survey there were two changes in benefit rates: one in March 1930, when benefit rates were changed for teenagers and adult dependants, and a general reduction in benefit rates in October 1931 towards the end of the period covered by our data. Benefits were available for relatively long periods of unemployment. Eligibility for benefit had to be established by a requirement (the First Statutory Condition) of 30 weeks of contributions within the previous two years. From April 1928, benefit could be received for up to 74 weeks after first serving a waiting period of six days on becoming unemployed. 6 Throughout the interwar period, there were supplementary schemes for those who failed to qualify for the standard insurance benefits or who had exhausted their entitlement. 7 Transitional Benefits introduced in 1928 were available at the same rates as the standard scheme for 96 weeks for those who had made eight contributions in the previous two years and indefinitely for those who had made 30 contributions at any time. The effect of the relatively relaxed conditions introduced in 1928 was, in the words of one observer, to provide almost unlimited benefits to those who could show very limited amounts of insurable employment (Burns 1941, p. 51). But these conditions were tightened in November 1931, with the restriction of standard benefit to 26 weeks in any benefit year and with the introduction of a new supplementary system, Transitional Payments, which was subject to a household means test. 8 Receipt of benefit was subject to certain conditions. Under the Third and Fourth Statutory Conditions, the applicant had to be available for work and genuinely seeking work. In March 1930 the genuinely seeking work clause was abolished. The numbers of unemployed on the register grew with the worsening depression, but some of the increase was attributed to the newly relaxed conditions for benefit. Under pressure to control expenditure, the government, as well as reducing rates of benefit and increasing contributions, introduced the Anomalies Regulations in October These were aimed at reducing successful claims from workers whose attachment to the labour force was thought to be marginal: they applied particularly to certain classes of part-time, seasonal and casual workers and to married women. 10 There have been conflicting views about the generosity of the system. Benjamin and Kochin argued that unemployment benefits were on a more generous scale relative to wages than ever before or since (1979a, p. 442). Although average benefit wage ratios were not as high as they suggested, 11 benefits were receivable for long periods once a successful claim was established. But it has also been argued that eligibility was tightly controlled,

4 634 ECONOMICA [NOVEMBER particularly through the genuinely seeking work clause. As Deacon (1976) has illustrated, a growing number of claims were disallowed during the 1920s, a trend that was reversed after March During the period covered by the bulk of our data ( ), fewer than 5% of claims were disallowed. 12 Hence, if benefits did have any disincentive effects, they should be observable during these years. II. THE NEW SURVEY OF LONDON LIFE AND LABOUR The NSLLL remains the only source of microdata for the interwar period that is usable for the analysis of unemployment incidence. The NSLLL was the largest of a series of social surveys undertaken in various towns and cities during the interwar period to identify the extent and causes of poverty among working-class households. The survey was undertaken at the London School of Economics under the direction of Sir Hubert Llewellyn-Smith between 1928 and 1932, although the bulk of the interviews took place during the years The purpose of the project was to assess working-class progress since Charles Booth s celebrated study forty years earlier. 13 Unlike Booth, however the interwar investigators undertook a detailed house survey which involved visiting randomly selected working-class households in 38 London boroughs, an area somewhat larger than that investigated by Booth. 14 The household survey itself was carefully conducted, and the chosen households were often visited repeatedly until the relevant information was obtained. Those subsequently found to be middle-class on the survey s definition were excluded, 15 as were Jewish households. The non-response rate was only 5% remarkably low by modern standards and, with the exception of a few variables, there is very little missing information on the completed cards. 16 Three types of information were included on the record cards: data on the household s demographic structure and family composition; information about occupation and earnings, and housing details, including rent paid. Because the survey was designed to compare household income with a scale of minimum needs in order to measure poverty, it carefully distinguished earners from nonearners. It documented in full their income from wages and other sources, together with work expenses such as travel-to-work costs and health and unemployment insurance contributions. The details of wages and employment are given both for the week preceding the date of the interview and for a full week. Information was also reported on the earners occupations and in some cases their employers occupations. We have now computerized the whole of the existing records from the survey a total of 26,915 households. 17 This, together with the coding of a wider range of variables, allows us to pursue some of the issues raised by Benjamin and Kochin but not fully addressed in Eichengreen s study. First, with only a 10% sample of the NSLLL data, it was not possible for Eichengreen to analyse young workers separately from adults, and because of their low participation rates, females were omitted altogether from the analysis. Thus, some of the key hypotheses put forward by Benjamin and Kochin could not be investigated in detail. With a much larger data-set we are able remedy this.

5 2002] UNEMPLOYMENT INCIDENCE IN INTERWAR LONDON 635 Second, Eichengreen s data-set did not contain variables on the individual s industry or occupation. Yet many observers of the interwar labour market would argue that such characteristics are vital to the understanding of differences in unemployment incidence. If such characteristics had an independent effect on unemployment and were also correlated (as one might expect) with the benefit wage ratio, then omitting them might affect the coefficient on the latter. As part of our reworking of the data, we have used the available information to provide a coding by skill and by occupational sector. This allows us to examine the effects of industry and skill in determining individual unemployment probabilities. Third, benefit rates changed during the survey period. Eichengreen used two alternative sets of rates in his estimates and the results seem to be sensitive to the set that was used. It is therefore important to establish which benefit regime applied to each case in the data. We have used the interview dates to distinguish between different benefit regimes and to allow for year effects. 18 III. EXPLAINING UNEMPLOYMENT INCIDENCE The NSLLL provides only a snapshot of the individual s labour market status. We have no information on the duration of unemployment the variable often analysed in connection with unemployment benefits. Any effects of benefit rates on unemployment incidence will therefore reflect the net impact on inflows and outflows to and from unemployment. In the interwar labour market, rates of turnover were high and durations were relatively low. 19 It has sometimes been argued that the unemployment insurance system induced repeat spells of unemployment as employers laid off workers for short periods in rotation. If this was the case, then it would be reflected in higher incidence, but not necessarily in longer unemployment duration. We follow Eichengreen in analysing unemployment incidence using probit analysis. The unemployment status of individual i, U i, depends on the benefit wage ratio (B=W) i, other observed characteristics, X i, and unobserved characteristics, e i, such that (1) U i ¼ f 1 ((B=W) i ; X i ; e i ): The individual s wage is a function of observed characteristics, Y i, and unobserved characteristics, v i, such that (2) W i ¼ f 2 (Y i ; v i ): The data-set used for analysis includes all those aged who could be identified as labour force participants, excluding the self-employed and those who were sick, incapacitated or on strike. Where a worker was unemployed, this was often mentioned explicitly on the survey card or could be inferred directly from the receipt of unemployment benefit or from other information. 20 Those who were working full-time, part-time or short-time in the week of the survey are treated as employed (in the last case even if some unemployment benefit was received). Our definition of unemployment includes those who were not covered by unemployment insurance or who, for other reasons, were not receiving benefit. It probably excludes a good deal of unregistered

6 636 ECONOMICA [NOVEMBER unemployment and a good many discouraged workers, especially among women. To estimate equation (1), we need wage rates and unemployment benefit rates both for the employed and for the unemployed. For the wage, we use weekly earnings in a full week because we have this information both for those who were employed and for those who were unemployed on the date of the survey. We choose weekly rather than hourly wages on the grounds that the weekly wage is more appropriate for comparison with weekly unemployment benefit. 21 However, if the unobserved component of the wage (v i in (2)) is correlated with the unobserved component of the unemployment probability (e i in (1)), then the coefficient on the benefit wage ratio will be biased. In constructing the benefit wage ratio, therefore, we use the predicted wage from an equation estimated for the log of earnings in a full week. The specification of the earnings function and the results obtained from it are discussed in the Appendix. Although many of the explanatory variables are common to the X and Y vectors in equations (1) and (2), identification is achieved through the inclusion of borough dummies in the wage equation and also through the restriction imposed on the benefit wage ratio in the equation for unemployment incidence. Since unemployment benefits were given at a flat rate for the entire period of unemployment, according to the individual s age and family circumstances, it is relatively straightforward to impute a weekly benefit amount to all workers in the survey. 22 Thus, for example, an adult male would receive 17s (85p) for himself, 7s (35p) for a non-working wife and 2s (10p) for each child. Details of family relationships within the household, distinguishing between earners and non-earners, allow us to assign the relevant dependants allowance to each individual. 23 There were changes in the rates for teenagers and for adult dependants in March 1930 and a general reduction of benefit rates took place in October 1931, towards the end of the period covered by our data. We use our knowledge of the interview dates to apply the relevant rate of benefit to each case. The only cases where zero benefit was assigned are where the worker was in an occupation not covered by the insurance scheme. These are identified as cases where no insurance contribution is recorded and where the worker s occupation was not among those classified as insurable. Skill and industry dummies are derived from information on individuals occupations, which we have classified according to the occupational codes given in the 1931 census. These three-digit codes are aggregated into 31 different occupational orders, which roughly correlate to industries, and are also aggregated into five skill levels based on the Registrar General s mapping of occupational codes into social classes. Cases where the information on occupation was incomplete or unavailable were placed in occupational order 31, other and undefined workers, and were assumed to be unskilled. There is no other information on education or training, but since most skills were learned on the job this is not likely to be a serious omission. Age and marital status are also included as explanatory variables for unemployment incidence. In a significant minority of cases, no age is entered but adult or something similar appears on the record card. We have created a separate dummy for these rather than exclude them from the data-set. In addition, we include a variable for homeownership. As with marital status, this

7 2002] UNEMPLOYMENT INCIDENCE IN INTERWAR LONDON 637 is assumed to be a proxy for labour market quality and is likely to be correlated with the individual s past wage and employment history. Home ownership was relatively uncommon among working-class households in the interwar period, although it can be readily identified in the survey. It should be noted, however, that we can identify only whether an individual lived in an owner-occupied dwelling, not whether the individual was the owner. We also include a variable reflecting household income other than that received by the individual (either earnings or benefit) relative to the poverty line. 24 We define minimum needs, based on the age and sex composition of the household, according to the scale put forward by Rowntree (1936). 25 Under the assumption that household income was pooled, other household income would be expected to raise the reservation wage of an unemployed worker. Conversely, lack of income from other members of the household would be expected to intensify the breadwinner s search for employment. IV. ADULT MALES AGED Benjamin and Kochin argued that by 1931 weekly benefits exceeded 50% of average weekly wages a statement that has been widely challenged. But they also argued that, because benefit rates were not linked to the individual s wage rate, some workers would have had very high benefit wage ratios, giving rise to severe disincentive effects (1979a, p. 455). However, in the absence of evidence for individual workers, they were unable to substantiate their claim. By contrast, in his detailed study of adult male workers in Greenwich in 1931, E. Wight Bakke concluded thus: The behaviour of the unemployed in searching for employment gives no evidence that the possibility of drawing Unemployment Insurance benefit has retarded the efforts of the unemployed to get back to work. It has removed the cutting edge of the desperation that would otherwise attend that search (Bakke 1933, p. 143). We analyse males aged separately from younger males since the latter deserve separate attention. Adult males aged 25 and over are more likely to have had responsibilities for wives, children or other dependants, and they account for the bulk of the household heads among the earners in our data. The differences in benefit rates among these workers arise only from differences in dependants allowances, unless they were ineligible for any benefit at all. Thus, it is predominantly among these that the rate of benefit taken as standard in so many studies applies: 24s ( 1.20) for the earner plus a wife and two children. Despite this, however, the average benefit wage ratio calculated for our data is only 0.42, rather lower than that suggested by Benjamin and Kochin. The frequency of different benefit wage ratios for all males aged (employed and unemployed) is plotted in Figure 1. These (and other descriptive statistics reported below) are for imputed rates of benefit relative to observed weekly earnings, rather than to the predicted earnings used in the regressions. The bulk of workers had benefit wage ratios in the range , and 77% had ratios less than the figure of 0.5 quoted by Benjamin and Kochin even though younger workers have been excluded. The few with rates over 1 are largely the result of unusually low wage rates. Eichengreen s study suggested that non-household heads were more susceptible than non-heads to benefit-induced unemployment. Following

8 638 ECONOMICA [NOVEMBER 6000 Imputed benefits/wages No. of individuals Imputed benefit wage ratio FIGURE 1. Imputed benefit wage ratios, males aged 25 and above. him, we distinguish these two groups in our data. Table 1 summarizes the means of the variables described in the previous section, distinguishing between heads and non-heads. The unemployment rate among household heads is strikingly low at 6.6% but it is over 10% for the non-heads. Yet the average benefit wage ratio is much lower for non-heads than for heads. This is largely because heads were much more often married with children. It is interesting also to compare the employed with the unemployed. For both groups, the benefit to wage ratios are only very marginally higher for the unemployed than for the unemployed. An aspect not previously explored for the interwar period is the difference in skill levels between the employed and the unemployed. For both heads and non-heads, more than half the unemployed were unskilled. By contrast, among the employed less than 30% were unskilled. There were fewer semi-skilled among the unemployed than the employed but, surprisingly, the proportion of skilled (the omitted group) is only slightly lower among the unemployed. These findings suggest that it might have been low skill levels rather than high benefit wage ratios that were associated with the incidence of unemployment. Notable too is the higher level of other household income among unemployed than employed heads and higher levels still among non-heads. In order to untangle these effects, we estimate probit equations (unemployment = 1; employment = 0) for household heads and non-heads. The benefit wage ratio is constructed from imputed benefit and predicted earnings. The results presented in Table 2 are for the marginal probabilities and the associated z statistics (probit coefficients=standard errors). The result for the household heads in the first column of the table exhibits the U shape of unemployment incidence by age well known from other studies with the minimum at around age 43. Between ages 25 and 45 the probability of unemployment declines by about 2 percentage points. 26 Being either married or living in an owner-occupied dwelling also reduces the probability of unemployment by about 3 percentage points. Other household income relative to minimum needs has a positive effect. Perhaps this is not surprising, given

9 TABLE 1 SAMPLE STATISTICS FOR MALES AGED Household heads Non-heads All Employed Unemployed All Employed Unemployed Unemployed(%) Imputed benefit wage ratio Semi-skilled (%) Unskilled (%) Age (years) Married (%) Owner-occupier (%) Other income=minimum needs Number 15,821 14, ] UNEMPLOYMENT INCIDENCE IN INTERWAR LONDON 639

10 640 ECONOMICA [NOVEMBER TABLE 2 PROBIT EQUATIONS FOR UNEMPLOYMENT INCIDENCE: MALES AGED a Household heads Non-heads (1) (2) (3) (4) Age (5.44) (4.76) (1.50) (0.49) Age (5.77) (5.30) (1.63) (0.78) Age not known (5.91) (4.95) (1.72) (0.51) Married (4.50) (7.41) (1.32) (0.74) Owner-occupier (3.04) (3.91) (0.07) (0.78) Other income=minimum needs (12.54) (13.77) (2.18) (1.87) Benefit wage ratio (1.37) (7.16) (0.88) (1.76) (4.11) (3.35) (2.42) (1.77) (0.04) (1.11) (0.88) (0.25) Semi-skilled (1.07) (1.51) Unskilled (3.89) (0.75) Industry dummies Yes No Yes No Pseudo R Log likelihood No. of observations 15,821 15, a z-statistics in parentheses. that these other sources of income amounted to only a third of minimum needs on average. Even with unemployment benefits, 73% of the households would have been in poverty if the head had been unemployed. However, the effect on unemployment is not large: adding 1 to other income (41% of average minimum needs) would increase unemployment among household heads by 1.5 percentage points. 27 The benefit wage ratio gives a small and insignificant positive coefficient. Increasing this ratio from, say, 0.25 to 0.5 would raise the probability of unemployment by only 1 percentage point. 28 This offers little support for widespread benefit-induced unemployment at least among mature household heads. While the semi-skilled exhibit a small and insignificantly higher probability of unemployment than the skilled (the excluded group), the effect is significant for the unskilled. The coefficient for the unskilled indicates a higher probability of unemployment equivalent to 2 percentage points. While this is fairly modest, it must be remembered that the regression also includes a set of dummies for occupational orders and that the effect of skill is somewhat greater when these are excluded. Finally, the year dummies indicate a significant negative effect for 1930 (relative to 1929) but not for Even

11 2002] UNEMPLOYMENT INCIDENCE IN INTERWAR LONDON 641 though the depression was milder in the South-east and the recovery started earlier there than in other regions, the very small effect for the 1931 dummy is somewhat surprising. The second column of the table excludes both the skill and the industry dummies. Most of the coefficients are similar to those in column (1); but that on the benefit wage ratio increases noticeably, and becomes highly significant. This coefficient implies that raising the benefit wage ratio from 0.25 to 0.5 would add 4.3 percentage points to the unemployment rate. This offers strong evidence that failure to control for skill and industry tends to bias the benefit wage coefficient upwards. It suggests that for unskilled workers higher unemployment incidence was due to their lack of skills rather than their low wages relative to benefits. The final two columns of Table 2 present the results for the same equations estimated for non-household heads. The coefficients are generally much less significant. The main differences are positive effects for married, and negative effects for other income relative to minimum needs. These coefficients are difficult to interpret, but suggest that household responsibilities might have weighed less heavily for non-household heads. It is interesting to note also that the coefficients on the benefit wage ratio are negative in both regressions. The effect of unemployment benefit is negative and insignificant, but, as for the older males, it becomes positive when skill and industry coefficients are omitted. The results in columns (2) and (4) differ slightly from those of Eichengreen, who found more significant effects for non-heads. However, most of the non-heads in Eichengreen s sample are relatively young and we have excluded the under-25 from our sample. 29 We turn to these next. V. MALES AGED One of the strongest claims made by Benjamin and Kochin was that the effects of differences in benefit wage ratios could be seen most clearly in the unemployment rates for young workers. Labour exchange statistics show that the unemployment rate for male juveniles was only a little over one-third of that for all males. Unemployment rates gradually rose with age to a peak in the early twenties. According to them, Both the low level of benefits and the lengthy period that was required to achieve eligibility for benefits made unemployment an unattractive prospect for juveniles, hence their low unemployment rate. At age 18 benefits increased sharply relative to wages and eligibility for supplementary schemes began, while benefits relative to wages increased again at age 21. These facts also are compatible with increases in unemployment. (Benjamin and Kochin 1979a, p. 459) Such evidence, though suggestive, is hardly compelling. The only further evidence offered in support was an insignificant coefficient on the benefit wage ratio in a time series equation for juvenile unemployment estimated for the years Although Benjamin and Kochin argued that juvenile unemployment was one of the few things that went right during the interwar years, others have seen it as a much more serious problem. 30 It was often argued that low unemployment rates for those aged reflected an abundance of dead-end

12 642 ECONOMICA [NOVEMBER jobs: jobs that provided little in the way of training and skills, that were insecure, and that led to low future employability. As Beveridge put it, They enter, not as learners, but as wage earners, doing some work too simple or light to require the services of grown people. When, therefore, they themselves grow up and begin to expect the wages of grown people, they must go elsewhere to obtain these wages. They leave or are dismissed and their places are taken by a fresh generation from the schools. They find themselves at eighteen or twenty without any obvious career before them, without a trade in their hands, and with no resource save unskilled labour. They go therefore very likely after an interval of military service to overcrowd that already crowded market. (Beveridge 1930, pp ) Beveridge wrote this originally in 1909, but investigations in the interwar period suggest it was just as relevant then as it had been before the First World War. 31 This fact alone suggests it is unlikely that rising unemployment with age is explained wholly by the age profile of unemployment benefit rates. Nevertheless it is possible that the interwar benefit system made it even more marked than it would otherwise have been. Under the schedule prevailing until March 1930, males could receive 6s (30p) at age 16 and 17, 10s (50p) at 18, 12s (60p) at 19, 14s (70p) at 20, and the full adult rate of 17s (85p) at 21. From March, 1930 the rates for 18 and 19 year olds were raised to 14s and the rate for 17 year olds to 9s (45p). 32 Although benefit rates clearly rose with age, it is not clear that benefit wage ratios did. Previous calculations have relied on aggregate wage data by broad age groups. Table 3 reports the average benefit wage ratio by year of age of the individuals in our data. This indicates that the average benefit wage ratio did indeed rise with age up to the age of 18, when it exceeded 40%. But, contrary to the received wisdom, from age 18 until 21 or 24 the benefit wage ratio actually declined. 33 As the lower row shows, unemployment incidence among the young males in our sample did rise sharply and continuously with age from 16 to 21. Over this age range, the unemployment rate tripled from 4.3% for the 16 year olds to 12.6% for the 21 year olds. Thus, the correspondence by age between benefit wage ratios and unemployment incidence is not as close as has been suggested. Sample statistics for these young males appear in Table 4. Only 7.5% of these were household heads and so we did not analyse heads and non-heads separately. The average benefit wage ratios are the same for the employed and the unemployed. For the whole sample, the proportions of semi-skilled and unskilled are slightly higher than for those aged 25 and over. But it should be noted that apprentices and trainees are allocated to the skill class of the occupation for which they were training; thus, an apprentice carpenter would be given the carpenters occupational code and would therefore be counted as TABLE 3 UNEMPLOYMENT AND BENEFITS BY AGE: MALES AGED Age Benefit wage Unemployment rate (%)

13 2002] UNEMPLOYMENT INCIDENCE IN INTERWAR LONDON 643 TABLE 4 SAMPLE STATISTICS FOR MALES AGED All Employed Unemployed Unemployed (%) Imputed benefit wage ratio Semi-skilled (%) Unskilled (%) Age (years) Married (%) Owner-occupier Other income=minimum needs Number skilled. Even so, it is surprising, given some of the comments in the literature, that the breakdown by skill is so similar to that for older workers. As with the older males, the unemployed were more often unskilled and less often semiskilled than the employed. Finally, and not surprisingly, few of these workers were married or were living in owner-occupier households. The average level of other household income relative to minimum household needs is similar to that for the older non-household heads. The results from probit estimates are presented in Table 5. Focusing on the first column, the age coefficients give a rising profile of unemployment incidence by age, reaching a peak at age 22 and then declining, but this accounts for an increase of less than 1% in the unemployment rate between the ages of 16 and 22. The effect of being married is negative and that of being a household head is positive, but neither coefficient is significant. The insignificant coefficients for owner-occupier households and for the ratio of other income to minimum needs give the same signs as those for older nonheads. The benefit wage ratio gives a negative coefficient on the borderline of significance. When industry and skill dummies are excluded, as in column (2), the coefficient becomes positive and significant. As with the older males, it seems that failure to allow for skill and industry generates spuriously positive benefit effects. Returning to column (1), the lack of any effect arising from being unskilled is surprising. It contrasts sharply with the result for household heads in Table 2, and with the finding in Table 3 that such a large proportion of the young unemployed were unskilled. It may simply be that the ease of obtaining unskilled jobs at 16 or 17 was offset by the difficulty of obtaining work for those remaining unskilled at slightly higher ages. This is what the dead-end jobs hypothesis implies, and in order to test this we interacted unskilled with age. The small positive and insignificant coefficient in column (3) suggests that this hypothesis has no explanatory power. It may be however, that the effect of dead-end jobs is being reflected in the industry coefficients rather than the skill coefficients. Column (4) excludes the industry dummies, and both the interaction term and the benefit wage ratio remain insignificant. Thus, neither the benefit hypothesis nor the dead-end jobs hypothesis seems capable of explaining much of the increase in unemployment incidence with age among young males.

14 644 ECONOMICA [NOVEMBER TABLE 5 PROBIT EQUATIONS FOR UNEMPLOYMENT INCIDENCE: MALES AGED a (1) (2) (3) (4) Age (3.79) (1.54) (3.76) (4.14) Age (3.52) (1.19) (3.51) (3.85) Married (0.29) (1.42) (0.29) (0.35) Household head (0.06) (0.62) (0.06) (0.38) Owner-occupier (1.63) (2.07) (1.63) (1.64) Other income=minimum needs (0.51) (1.35) (0.50) (0.84) Benefit wage ratio (1.96) (3.65) (1.96) (0.79) (1.26) (0.54) (1.26) (0.57) (0.03) (0.91) (0.03) (0.78) Semi-skilled (1.74) (1.73) (0.20) Unskilled (1.57) (0.24) (0.82) Unskilled age (0.03) (0.86) Industry dummies Yes No Yes No Pseudo R Log likelihood No. of observations a z-statistics in paretheses. VI. FEMALES Benjamin and Kochin argued that the unemployment insurance system influenced the behaviour of females in two ways: first by causing some to substitute unemployment for employment, and second by inducing some who would otherwise have been out of the labour force to register as unemployed. They suggested that virtually all of the insurance induced unemployment among men and single women represented substitution out of work (1979a, p. 464). But for married women the issue was more one of easy access to benefit, inflating the measured labour force. They pointed in particular to the effects of changes in the regulations determining eligibility for benefit under the Anomalies Regulations, which were introduced in October Most of the argument, and evidence, related to the effects of the Anomalies Regulations in reducing the number of women registering as unemployed. Most of the NSLLL interviews occurred during the time of relatively relaxed conditions prevailing before the Anomalies Regulations were introduced. But the rate of unemployment recorded for women in the survey is remarkably low. For all women aged 14 and over, the rate of unemployment

15 TABLE 6 SAMPLE STATISTICS FOR FEMALES Aged Aged All Employed Unemployed All Employed Unemployed Unemployed (%) Imputed benefit wage ratio Semi-skilled (%) Unskilled (%) Age (years) Married (%) Owner-occupier Other income=minimum needs Number ] UNEMPLOYMENT INCIDENCE IN INTERWAR LONDON 645

16 646 ECONOMICA [NOVEMBER is just 2.9% compared with 6.5% for the same boroughs (and the same age group) at the Census of April In part this reflects the fact that employment conditions were worse at the time of Census than they were, on average, over the survey period. 34 But it also seems likely that many women who declared an occupation to the census takers were effectively out of the labour force and were not identified as labour market participants in the NSLLL. 35 The women included as earners in our data are therefore those who are most likely to have been influenced by the incentive effects of unemployment insurance benefits. As with the males, we divide the female earners in our sample into those aged and those aged 25 and above. Table 6 provides summary statistics. For the older women, average benefit wage ratios exceed 0.5 and are identical for the employed and the unemployed. For the younger workers, average benefit wage ratios are lower and fairly close to the rates found for men but they are slightly higher for the unemployed than for the employed. Like the men examined earlier, both groups of women show much higher proportions of unskilled among the unemployed than employed. About a quarter of the older women were married and their average age is relatively low, while very few of the younger women were married. The high replacement rates for the older women deserve further attention. The benefit rate for a female aged 21 or over was 15s (75p), compared with 17s (85p) for men. Although men more often received allowances for dependants, this is more than outweighed by the lower wage rates for women. Some of this reflects part-time work, but part-time workers make up only about 7% of the sample. Figure 2 indicates that, while a significant minority had benefit wage ratios of zero, about a third had benefit wage ratios exceeding 0.6. The few with very high benefit wage ratios, again, reflect unusually low wage rates. Probit estimates for women aged appear in the first two columns of Table 7. In column (1) the age coefficients are insignificant, suggesting that the U shaped profile of unemployment incidence by age that was so prominent for men aged 25 and older is absent for women. The negative coefficient on married indicates that married women were less likely than single women to be unemployed. This contrasts with much higher unemployment rates among married than single women in the Census and it suggests that married women who were unregistered or uninsured were unlikely to appear as unemployed in the NSLLL. 36 The benefit wage ratio gives a small negative and insignificant coefficient. Hence there is no evidence of benefit-induced unemployment for mature females. 37 As for males, the skill coefficients imply higher unemployment incidence for semi-skilled and unskilled than for skilled. When skill and industry dummies are excluded, as in column (2), the coefficient on the benefit wage ratio remains small and highly insignificant. Perhaps it is worth further stressing that the women counted as unemployed in the NSLLL are those most likely to have been affected by unemployment insurance: had more women on the margins of the labour force been included, it seems even less likely that any positive effect would be found. Like their male counterparts, young females faced benefit rates that were graduated by age. Under the rates prevailing up to March 1930, they received 5s (25p) at 16 17, 8s (40p) at 18, 10s (50p) at 19, 12s (60p) at 20 and the full

17 2002] UNEMPLOYMENT INCIDENCE IN INTERWAR LONDON No. of individuals Imputed benefit wage FIGURE 2. Imputed benefit wage ratios, females aged 25 and above TABLE 7 PROBIT EQUATIONS FOR UNEMPLOYMENT INCIDENCE: FEMALES a Age Age (1) (2) (3) (4) Age (0.02) (0.22) (0.54) (1.42) Age (0.12) (0.27) (0.55) (0.21) Age not known (0.25) (0.01) Married (1.55) (1.75) (0.04) (0.12) Household head (1.55) (0.58) Owner-occupier (1.00) (1.31) (0.97) (1.42) Other income=minimum needs (0.36) (0.16) (0.84) (1.46) Benefit wage ratio (1.10) (0.24) (0.84) (2.39) (0.21) (0.28) (0.01) (0.21) (0.53) (0.46) (0.06) (0.11) Semi-skilled (2.07) (2.57) Unskilled (3.12) (4.86) Industry dummies Yes No Yes No Pseudo R Log likelihood No. of observations a z-statistics in parentheses.

18 648 ECONOMICA [NOVEMBER TABLE 8 UNEMPLOYMENT AND BENEFITS BY AGE: FEMALES AGED Age Benefit wage Unemployment rate rate of 15s (75p) at 21. After March 1930 rates were raised to 7s 6d (37.5p) for 17 year olds and to 12s (60p) for 18 and 19 year olds. Very few received dependants allowances. Table 8 reports benefit wage ratios by age. The slight decline between ages 18 and 20 reflects the fact that after March 1930 the benefit rate was constant across these ages while wages increased with age. But from age 17 until 21 there is a larger increase the benefit wage ratio (from 0.34 to 0.53) than there was for young males. This steeper rise for females reflects the fact that wages rose less rapidly over this age range (a fact that is consistent also with the high rates for females 25 and over). At first sight it might seem that this rise is consistent with Benjamin and Kochin s observations about young workers mentioned earlier. But, as the lower row of Table 8 shows, there is essentially no increase in unemployment incidence with age for females. This indicates the importance of distinguishing between young males and young females: benefit wage ratios rose most sharply for females, while unemployment incidence rose most sharply for males. The probit results for young females appear in columns (3) and (4) of Table 7. In column (3) there is no evidence of the inverted U of unemployment incidence by age as there was for young males. The effect of the benefit wage ratio is marginally positive and very insignificant, but the dummy for semi-skilled now gives a negative coefficient, probably reflecting the buoyancy of employment in assembly-line jobs. When skill and industry dummies are excluded from the equation the coefficient on the benefit wage ratio becomes significant although the coefficient remains very small. Thus, although benefit wage ratios increased sharply over the age range, there is still no evidence of a substantial benefit effect on the incidence of unemployment. VII. CONCLUSION Benjamin and Kochin argued that the army of the unemployed in the interwar period was largely a volunteer army, but Eichengreen concluded from his analysis of microdata from London that the volunteers were a mere squadron. Our results suggest that the effects of the benefits on unemployment were even smaller than this. However, when skill and industry dummies are excluded, significant positive benefit effects do emerge. This suggests that unemployment incidence was associated with the individual s skill and industry and that failure to allow for such effects leads to spuriously positive effects for the benefit wage ratio. We have presented for the first time estimates for young males and for women, groups singled out by Benjamin and Kochin as demonstrating the impact of benefit-induced unemployment. In none of these cases do we find significant positive effects of the benefit wage ratio on the incidence of unemployment once skill and industry effects are taken into account. For

19 2002] UNEMPLOYMENT INCIDENCE IN INTERWAR LONDON 649 young workers, even the raw data paint a rather different picture than was previously suggested. Benefit wage ratios did not rise much with age for males, and unemployment incidence did not rise with age for females. Finally, even for adult women, the group with the highest benefit wage ratios and about whom the policy-makers were most concerned there is no evidence that benefits induced unemployment. APPENDIX: EARNINGS FUNCTIONS FOR WORKERS IN THE NSLLL This appendix describes the earnings functions estimated for men and women on the data for a full week s earnings in the NSLLL. These are estimated for two purposes: to generate the variable for the residual wage used in our probit models for unemployment, and to fill in the full week s wage in the small proportion of cases where this information was missing. The earnings function takes the log of full week s earnings as the dependent variable. Since we do not have variables for education and experience, we rely instead on age and skill category. There are two slightly unusual features to note about the model. First, by inspection of the data, it is clear that earnings do not follow a simple quadratic function of age. This feature has been identified in several recent studies of TABLE A1 WAGE EQUATIONS FOR MEN AND WOMEN AGED a Men Women Constant (100.15) (63.07) Age (31.18) (18.05) Age (23.83) (15.36) Age (9.05) (3.49) (Age-25) (23.37) (15.15) Age not known (41.95) (21.66) Professional (5.11) Managerial (5.45) (4.03) Skilled (3.47) (1.89) Semi-skilled (2.103) (1.21) Unskilled (0.68) (0.015) Industry dummies Yes Yes Borough dummies Yes Yes R RSS No. of observations 22, a Dependent variable: Log full week s wage. t-statistics in parentheses.