and 3 on employment and unemployment histories. The number of spells .... Table 1 shows the frequency distribution of unemployment spells for men entering the .... 23% from education or training and 7% from long-term sick. The increase in ...
THE CHANGING PICTURE OF MALE UNEMPLOYMENT IN BRITAIN
Mark P Taylor and Alison L Booth
ESRC Research Centre on Micro-social Change University of Essex Wivenhoe Park CO4 3SQ
October 1996
ABSTRACT This paper uses panel and retrospective life history data from an important new data source - the British Household Panel Survey - to establish some stylised facts about the unemployment experiences of men. In particular we investigate the proportion of the sample who suffer from repeated unemployment spells, the origin and destination states of unemployment spells, some reasons for entering unemployment, and the length of time spent unemployed over different periods and across labour market entry cohorts. JEL Classification: I21, J24, J42. Keywords: unemployment incidence, unemployment duration
Acknowledgements: We should like to thank Nick Buck for helpful discussions about the data. Any errors remain ours. The support of the Economic and Social Research Council, the University of Essex and the Leverhulme Trust is gratefully acknowledged.
The Changing Picture of Male Unemployment Non-Technical Summary According to the British Household Panel Survey, over forty percent of adult British men have experienced some unemployment over their working lives, while nearly twenty percent have experienced multiple unemployment spells. How and why do men enter unemployment, and where do they go after unemployment? For how long do periods of unemployment last? Have there been changes in the pattern of unemployment over calendar time and across cohorts? This paper answers some of these questions, and provides some stylised facts about the changing unemployment experiences of British men, by using panel and retrospective lifetime employment data collected by the British Household Panel Study in successive years since 1991. Aggregate unemployment rates conceal the distribution of unemployment across the labour force and across individuals over time. They tell us nothing about the duration of unemployment spells, or about repeated spells. Retrospective and/or panel data on individuals’ working lives are needed in order to obtain information on these kinds of issues. The data source used here, the British Household Panel Survey (BHPS), is a nationally representative survey of some 5,500 households (and approaching 10,000 individuals) randomly selected south of the Caledonian Canal. We choose the BHPS for our analysis since, not only does it contain data regarding employment status at each interview and for the (approximately) 12 month period between interviews, but also detailed retrospective information collected at Waves 2 and 3 on employment and unemployment histories. The number of spells of unemployment experienced in early years of the work history for older members of the workforce in the BHPS may, due to recall problems, be an under-estimate of the true number, or may simply be a measure only of longer spells of unemployment that are more likely to be remembered by respondents. Because of these potential problems of under-reporting of unemployment occurrences in the more distant past, we use a number of different cohorts and calendar periods in order to answer the questions posed at start of this paper. We look, for example, at sub-samples of men entering the labour market for the first time since 1914; after World War Two; after the first oil price shock of 1972; and after Thatcher came to power in 1979. While some of these work history data may be subject to problems of recall, there is no reason to expect these biases to vary across characteristics other than age, and anyway, they are better than no data at all. We also focus on the number of unemployment spells experienced since September 1990, thereby using the panel data from the first four waves of the BHPS, in which respondents are asked to detail labour market movements during the 12 month period before each interview. The average number of unemployment spells experienced by men in the BHPS sample is 0.8, with 56% of men not experiencing any unemployment at all. The data, however, reveals some important trends. More recent labour market entry cohorts have experienced more unemployment spells on average than earlier cohorts (the mean rises to more than one spell per man for those entering the labour market after 1972), and are more prone to repeated spells. These spells also tend to be shorter in duration, although individuals actually spend a greater proportion of their time unemployed. More recent cohorts are also less likely to experience redundancy, but suffer more dismissals and non-renewed temporary work. As well as comparing labour market entry cohorts, the paper also contrasts the experiences of the male sample as a whole with those experienced by men currently 2
active in the labour market since 1990. This shows that those currently active in the labour market are more likely to have moved between unemployment and selfemployment than in previous years, and are less likely to have entered unemployment voluntarily. This is perhaps a reflection of the changes in the labour market that have occurred in the last fifteen years.
3
I. Introduction According to the British Household Panel Survey, over forty percent of adult British men have experienced some unemployment over their working lives, while nearly twenty percent have experienced multiple unemployment spells. How and why do men enter unemployment, and where do they go after unemployment? For how long do periods of unemployment last? Have there been changes in the pattern of unemployment over calendar time and across cohorts? This paper answers some of these questions, and provides some stylised facts about the changing unemployment experiences of British men, by using panel and retrospective lifetime employment data collected by the British Household Panel Study in successive years since 1991. Aggregate unemployment rates conceal the distribution of unemployment across the labour force and across individuals over time. They tell us nothing about the duration of unemployment spells, or about repeated spells. Retrospective and/or panel data on individuals’ working lives are needed in order to obtain information on these kinds of issues. The data source used here, the British Household Panel Survey (BHPS), is a nationally representative survey of some 5,500 households (and approaching 10,000 individuals) randomly selected south of the Caledonian Canal1. The first wave of the BHPS was conducted from September to December 1991, with subsequent waves collected annually thereafter
(see
Taylor, 1996, for detailed
information about the survey). We choose the BHPS for our analysis since, not only does it contain data regarding employment status at each interview and for the (approximately) 12 month period between interviews, but also detailed retrospective information collected at Waves 2 and 3 on employment and unemployment histories. Both the work history data and the wave-on-wave data from the BHPS are analysed in this paper. The work history data may suffer from problems of recall, in spite of the fact that the BHPS work history questionnaire was carefully designed to minimise such problems. For example, respondents are first asked about marriage and fertility histories, since marriage and birth dates are unlikely to be forgotten, and then asked about their work histories. Hence a chronological ordering of personal histories is developed, aiding the recall of employment events. Nonetheless, even with such prompting, individuals may be unable to remember with any degree of precision the details of unemployment spells experienced many years ago. The number of spells of
4
unemployment received in early years of the work history for older members of the workforce in the BHPS may therefore be an under-estimate of the true number, or may simply be a measure only of longer spells of unemployment that are more likely to be remembered by respondents. Because of these potential problems of under-reporting of unemployment occurrences in the more distant past, we use a number of different cohorts and calendar periods in order to answer the questions posed at start of this paper. We look, for example, at sub-samples of men entering the labour market for the first time since 1914; after World War Two; after the first oil price shock of 1972; and after Thatcher came to power in 1979. While some of these work history data may be subject to problems of recall, there is no reason to expect these biases to vary across characteristics other than age, and anyway, they are better than no data at all2. We also focus on the number of unemployment spells experienced since September 1990, thereby using the panel data from the first four waves of the BHPS, in which respondents are asked to detail labour market movements during the 12 month period before each interview. We construct unemployment spell variables using both the lifetime unemployment histories collected at Wave 2 and the one-year employment histories collected at each wave. The lifetime histories are used to construct the number of unemployment spells prior to September 1990, and subsequent spells are recorded using the year-on-year work histories. By adding these numbers together, we are able to calculate the number of unemployment spells each individual has experienced up to and including the date of the Wave 4 interview. We also construct a variable counting the number of unemployment spells experienced since September 1990. This variable depends only on recalling labour market events in the year prior to each interview, and is therefore less likely to suffer from error. Classification into the state of unemployment relies on respondents' own definition, which may differ across individuals, and potentially over time. This may result in an under-estimate of the number of unemployment spells, as either only the longer spells are recalled, or because some spells may be remembered by respondents
1
Thus the north of Scotland is excluded. If better educated workers are more likely to recall unemployment spells and better educated workers experience fewer spells ceteris paribus, then there may be sample-selection biases. Similarly, if those from earlier birth cohorts who experienced more unemployment spells are less likely to have survived until the 1990s, sample selection biases will occur.
2
5
as "out-of-the-labour force" as opposed to unemployment. Despite these potential drawbacks, we argue that these data provide us with important information on the changing nature of unemployment among British men in the latter half of the twentieth century. Moreover, the BHPS contains important information about the origin and destination states of unemployment spells, and about whether an unemployment spell has been entered into voluntarily or involuntarily. Section II of the paper examines the frequency distribution of unemployment occurrences, and shows the labour market origin and destination states of unemployment spells. Section III investigates the reasons given for leaving the job immediately preceding an unemployment spell, and is therefore able to shed light on the extent of voluntary and involuntary entry into unemployment. Section IV summarises the duration of unemployment spells, and charts changes in unemployment duration across time. The final section summarises the principal stylised facts emerging from our analysis. Throughout, unemployment is investigated for a number of different subsamples, and we are therefore able to map changes in unemployment across time.
II Unemployment Incidence Table 1 shows the frequency distribution of unemployment spells for men entering the labour market at, or after, the various calendar dates shown at the top of each column (with column percents in brackets). The sample in all cases includes men who were aged at least 16 years at the date of the Wave 1 interview in 1991 and who were fully interviewed at each of the first four waves of the BHPS. The final column shows the number of unemployment spells experienced after September 1990 by all male panel members of the sample aged between 16 and 55 and not in full-time education at Wave 1, irrespective of their date of entry into the labour market. From Table 1, it can be seen that the earliest year of entry into the labour market for a male sample member was 1914. Of this group, 56% have not experienced any unemployment at all over their entire lifetimes to date, while 96% have experienced fewer than four unemployment events. There appear to be very few men who frequently move into and out of unemployment; the average number of spells is just 0.79.
6
Table 1: Frequency distribution of the number of unemployment spells for men Number of spells 0 1 2 3 4 5 6 7 8 9 Mean Std Dev. N
Sample period 1914(56.4) 1808 (24.1) 774 (10.6) 341 (4.7) 150 (2.3) 74 (1.0) 37 (0.4) 12 (0.2) 5 (0.2) 5 (0.0) 1 0.793 1.205 3207
19461290 (51.2) 653 (25.9) 308 (12.2) (5.4) 137 (2.8) 71 (1.5) 37 (0.5) 12 (0.2) 5 (0.2) 5 (0.0) 1 0.915 1.283 2519
1973533 (44.9) 319 (26.9) 172 (14.5) (7.2) 85 (3.7) 44 (1.7) 20 (0.7) 8 (0.3) 3 (0.3) 3 1.084 1.364 1187
1980304 (41.1) 213 (28.8) 109 (14.7) (7.7) 57 (4.6) 34 (1.6) 12 (0.7) 5 (0.4) 3 (0.3) 2 1.171 1.401 739
19901644 (74.7) 365 (16.6) (5.8) 127 (2.2) 49 (0.5) 12 (0.2) 5
0.381 0.772 2202
The third column in Table 1 concentrates on the 2519 men entering the labour force after World War Two. Of these, 51% have experienced no unemployment to date, while 95% have experienced fewer than four unemployment spells. The average number of spells is, at 0.92, slightly higher than for the male sample as a whole. The fourth column in Table 1 focuses on men entering the labour market after the oil price shock of 1972. The percentage of men experiencing no unemployment spells falls to 45%, while the mean number of unemployment spells per person is also higher, at just over one spell. For men entering the labour market since 1980, only 41% have yet to experience any unemployment, while the average number of spells has risen further to 1.17. The increasing average number of unemployment spells for more recent ‘cohorts’ suggests either that more recent (younger) cohorts are more prone to unemployment, or that older cohorts under-report unemployment owing to recall error3. It is difficult to disentangle these two effects as both work in the same direction. The final column of Table 1 focuses on unemployment spells experienced since September 1990 by men aged between 16 and 55 and not in full-time education at Wave 1. This shows that 75% of this sample have experienced no unemployment since 1990, while only 9% have experienced multiple spells. Table 2 shows the labour market states from which unemployment spells were entered. Looking at the trends across different labour market entry cohorts, it is 3
Throughout the paper, the term ‘cohort’ is used to refer to particular sub-samples. Thus the 1980 ‘cohort’ refers to men entering the labour market after 1980.
7
interesting to note the decline in the proportion of unemployment spells that follow periods of employment, and the rise in the proportions following education/training and long-term sick. Of the unemployment spells experienced by the male sample as a whole (those entering the labour market since 1914), 77% originated from either employment or self-employment, 15% from education or government training schemes, and 3.5% from long-term sick. Of those entering the labour market since 1980, 68% of unemployment spells originated from employment or self-employment, 23% from education or training and 7% from long-term sick. The increase in the proportions entering unemployment from education can possibly be explained in two ways. Firstly, it may reflect the increased difficulty school leavers and graduates have in finding employment. Instead of entering work immediately upon leaving education, young men are increasingly having to spend time unemployed and searching for suitable employment; this now applies to graduates as well as other school leavers. Alternatively, it may merely reflect the fact that more recent cohorts have had less opportunity to find a job and then lose it, and therefore less opportunity to enter unemployment from employment or self-employment. Table 2: States from which unemployment spells originate Previous labour market state Employment Self-employment Education/Training Retired Long term sick Other Total number of spells
Sample period 1914-
1946-
1772 (69.9) 182 (7.2) 391 (15.4) 23 (0.9) 89 (3.5) 79 (3.1) 2536
1609 (70.0) 171 (7.4) 366 (15.9) 11 (0.5) 77 (3.3) 64 (2.8) 2298
1973824 71 294
1980-
(64.3) (5.5) (22.9)
528 36 191
(63.5) (4.3) (23.0)
64 (5.0) 29 (2.3) 1282
57 20
(6.9) (2.4) 832
1990573 (69.6) 108 (13.1) 66 (8.0) 5 (0.6) 28 (3.4) 43 (5.2) 823
The final column of the table supports this, showing that for men aged 16-55 at Wave 1 and not in full-time education, only 8% of unemployment spells originate from education, while 13% originate from self-employment. Table 3 shows the states that follow unemployment spells for each cohort/period that we are considering. The proportion of spells ending with employment remains reasonably constant across the labour market entry cohorts (at about 75%), while the proportion ending with self-employment has fallen from 8.5% for the earliest cohort to 4.6% for those entering the labour market since 1980. This is
8
likely to reflect an age effect, as self-employment tends to attract more mature workers (see, for example, Taylor, 1996, Campbell and Daly, 1992). Similarly, the increase in the proportion exiting to education and training may be explained by temporary or short spells of unemployment experienced by youth workers prior to or in between periods of education and/or training. Table 3: States to which unemployment spells terminate Next labour market state Employment Self-employment Education/Training Retired Long term sick Other Total number of spells
Sample period 1914-
1946-
1734 (74.7) 198 (8.5) 183 (7.9) 74 (3.2) 56 (2.4) 76 (3.3) 2321
1579 (75.2) 190 (9.1) 180 (8.6) 24 (1.1) 54 (2.6) 72 (3.4) 2099
1973907 77 137
1980-
(77.3) (6.6) (11.7)
597 36 112
(76.7) (4.6) (14.4)
31 (2.6) 21 (1.8) 1173
25 8
(3.2) (1.0) 778
1990470 (69.0) 88 (12.9) 54 (7.9) 9 (1.3) 22 (3.2) 38 (5.6) 681
The fifth column of Table 3 focuses on men aged 16-55 at Wave 1 who were not in full-time education. Comparing this with the other columns, it is clear that a smaller proportion post-1990 exit unemployment to employment (69%) and to education and training (7.9%), and a larger proportion exit to self-employment (12.9%). This supports our earlier suggestions about age effects.
III Voluntary and Involuntary Unemployment Unemployment incidence, or entry into the state of unemployment, may be either voluntary or involuntary. The relative costs of unemployment would depend upon its cause. One might expect the relative cost of voluntary unemployment to be rather low, since the decision to become unemployed was taken by the individual and hence one assumes utility flows are to be maximised by doing so. The BHPS data allows us to analyse the reasons for men entering unemployment, and how these have changed over time and across cohorts. For each spell of employment, respondents are asked to specify why they left. Thus, we not only know from which state they entered unemployment but, for those entering unemployment from employment, we also know precisely why the preceding employment spell was terminated. Unfortunately, questions about lifetime employment histories (asking for start and end dates of each spell of employment, self-
9
employment, unemployment, retired and other non-work activity experienced since leaving full-time education for the first time) and lifetime job histories (asking for details of each job experienced) were asked at separate waves (Waves 2 and 3). Thus, in order to match each unemployment spell to the details of the previous spell of employment, it is necessary to use information relating to the start and end dates of spells. Of course, if all sample members had perfect (or even consistent) memories, then this would not present a problem. In reality, however, there is a degree of recall error reflected in conflicting dates reported at each of the two waves that lifetime information was collected. Hence, Tables 4 and 5 below, showing the reasons given for leaving the employment spell immediately prior to an unemployment spell, include only spells where the month and year of leaving the employment spell match exactly with the month and year of starting the subsequent unemployment spell. Tables A1 and A2 in the Appendix relax this, to show reasons given for leaving an employment spell that ended within 12 months of a reported unemployment spell, to increase the sample sizes and provide a comparison.
Table 4: Reasons given for leaving job immediately preceding an unemployment spell Reason 1914Different/better job Redundancy Dismissed Temporary job ended Retired Health reasons Family care Other Total number of spells
195 535 97 176 18 41 8 189
(15.5) (42.5) (7.7) (14.0) (1.4) (3.3) (0.1) (15.0)
1259
1946190 486 91 166 10 35 7 179
(16.3) (41.8) (7.8) (14.3) (0.9) (3.0) (0.1) (15.4)
1164
Sample period 1973-
1980-
112 213 62 113
(18.2) (34.7) (10.1) (18.4)
53 117 47 82
(14.0) (30.9) (12.4) (21.6)
14 4 96
(2.3) (0.1) (15.6)
9 2 69
(2.4) (0.1) (18.2)
614
379
199019 210 37 90 6 13 3 85
(4.1) (45.4) (8.0) (19.4) (1.3) (2.8) (0.6) (18.4)
463
Concentrating initially on the first 5 columns of Table 4, that is the different labour market entry cohorts, a number of observable trends emerge. Firstly, there is a clear decline in the proportion of unemployment spells caused by redundancy, from over 42% of unemployment spells preceded by employment for those entering the labour market since 1914 to below 31% for those entering in 1980 and later. The proportion of unemployment spells caused by dismissals has risen across cohorts from 7.7% to 12.4%, while the proportion caused by the ending of temporary jobs has
10
increased from 14% for early cohorts to approaching 22% for the most recent cohort. This suggests that either younger workers are less likely to be made redundant, but are more likely to be dismissed or to be in temporary employment, or that more mature members of the sample recall job terminations rather differently from what actually happened (i.e. men ‘remember’ being made redundant, when in fact they were dismissed or in temporary employment). Table 5: Voluntary or involuntary separation from previous employment Reason 1914Voluntary Involuntary Total number of spells
410 849
(32.6) (67.4)
1259
1946386 778
(33.2) (66.8)
1164
Sample period 1973212 402
(34.5) (65.5)
614
1980124 255
(32.7) (67.3)
379
1990113 350
(24.4) (75.6)
463
By aggregating the reasons given for leaving the immediately preceding job, a variable can be constructed indicating the proportion of unemployment spells caused by voluntary and involuntary job terminations. Table 5 does this by collapsing the categories in Table 4. Redundancy, dismissals, temporary jobs ending and health reasons have been classified as involuntary job separations for this purpose, while leaving for a better or different job, retired, caring and other reasons have been classified as voluntary separations. Table 5 shows that the proportion of men entering unemployment voluntarily has actually remained relatively stable across labour market entry cohorts, at between 32.5% and 34.5%. However, the proportion who have entered voluntarily since 1990 is noticeably lower, at under 25%. This may reflect a reduced willingness to enter unemployment to search for more suitable employment in recent years due to increased competition for jobs or, alternatively, another instance of time distorting individuals’ perspectives of events4. IV. Unemployment Durations The welfare of the unemployed is likely to be closely related to unemployment duration, making the average duration of unemployment spells as important as absolute numbers. Long unemployment spells can result in loss of skills, reduced job search intensity, low self-esteem and self-confidence and a reduced attractiveness to
11
potential employers. Studies by Jackman and Layard (1991) and Van den Berg and Van Ours (1994) have suggested that, in Britain, the probability of leaving unemployment declines as unemployment duration increases. Arguments continue regarding whether this is caused by individual heterogeneity or duration dependence (e.g. Hildreth et al, 1995). The BHPS data allow estimates of unemployment spell duration to be constructed using the reported start and end dates of each spell, and these can be aggregated across spells to give estimates of the length of time an individual spends unemployed in total5. Table 6: Unemployment durations (months)6 Unit of observation
1914Spell Individual unemployed once or more Individual
Sample period 19461973-
1980-
1990-
11.21
10.39
8.46
8.25
7.35
(18.44)
(16.33)
(11.63)
(11.22)
(7.12)
19.66
18.75
15.84
15.58
10.13
(28.80)
(27.26)
(19.79)
(19.11)
(9.17)
8.10
8.63
8.37
8.69
1.86
(20.86)
(20.71)
(16.41)
(16.23)
(6.03)
Table 6 shows the mean unemployment duration in months for each unemployment spell, for each individual experiencing at least one unemployment spell, and for each individual regardless of whether an unemployment spell has been experienced. The mean length of an unemployment spell has clearly fallen for more recent cohorts, from a little over 11 months for the sample as a whole, to 8.25 months for those entering the labour market since 1980. The mean is reduced a further month by focusing on spells occurring after 1990 for men aged 16-55 and not in full-time education at Wave 1 of the BHPS. This is either a further reflection of recall error, with earlier (older) cohorts only remembering longer spells of unemployment, or evidence suggesting that unemployment spells suffered by men in Britain today are considerably shorter than those experienced by previous generations7.
4
Nickell (1980), using the 1972 General Household Survey, shows that 59% of unemployed men left their last job for redundancy or because they were sacked, and 4.5% left because a temporary job ended. Some 21% left employment voluntarily. 5 It should be remembered that the start and end dates are likely to be prone to recall error, despite the measures adopted in collection of the data. 6 Standard deviations in brackets. 7 The durations given in tables of mean durations focus on completed spells only. Thus, we might expect the later cohorts to have shorter mean durations as a larger proportion of longer spells will be censored.
12
A similar pattern is evident when looking at the total time individuals experiencing at least one unemployment spell have spent unemployed. This total has fallen from an average of approaching 20 months for the male sample as a whole to under 16 months for those entering the labour force since 1980. However, although the length of time spent unemployed has fallen in absolute terms, it has increased relative to labour market experience. Those entering the labour market since 1914 have had potentially their whole working lives (or about 50 years) in which to experience an average of 20 months unemployed, averaging out to about one month unemployed to every 2.5 years in the labour market. However, those entering the labour market since 1980 have had only a maximum of 15 years in which to experience an average of over 15 months of unemployment. This averages out to one month unemployed to every year in the labour market. This ratio is even greater for the post 1990 sample. It would appear, then, that more recent cohorts are spending a greater proportion of their time unemployed than older cohorts. Again, however, this is consistent with the effects of recall error, with older cohorts not remembering shorter unemployment spells experienced many years ago. To further investigate unemployment spell duration, Table 7 below presents mean unemployment spell duration by previous labour market state, while Table 8 disaggregates mean unemployment spell duration by subsequent labour market state. Concentrating initially on the second column of Table 7, the longest unemployment spells on average originate from retirement (at 12 months), followed by paid employment (11.6 months). The shortest spells on average originate from periods of long-term sick, at about 8 months. The downward trend in average durations is again noticeable across the groups as more recent cohorts are selected, with the exception of unemployment spells originating from other non-specified states. For the 1980 cohort, the longest unemployment spells originate from periods of education and training (at an average of 10.2 months), which may be caused by school leavers, while again the shortest originate from long-term sickness (at an average of 5 months) and also from self-employment (at 6.28 months). For the post 1990 sample, unemployment spells following retirement have the longest average duration, followed by self-employment, while those following long-term sickness and employment have the shortest average duration at about 7 months. Thus it would appear that unemployment spells originating from long-term sickness consistently have the shortest average duration
13
across cohorts and time periods, while those following retirement have the longest. The former perhaps is contrary to intuition; one might expect those entering unemployment from long-term sickness to struggle to find employment. However, further analysis has shown that over 10% of those who enter unemployment from long-term sickness actually exit to long-term sickness, while more than 30% exit to non-employment states. This is almost double the average proportions for the sample as a whole as described earlier in Table 3, and suggests that the long-term sick do have problems entering employment. The latter can perhaps be explained by age, those exiting retirement and becoming unemployed are likely to be aged over 50, and thus are less attractive to potential employers with a limited working life in front of them. Table 7: Unemployment duration (months) by previous (origin) labour market state Previous labour market state Employment Self-employment Education/Training Retired Long term sick Other Total
Sample period 1914-
1946-
1973-
1980-
1990-
11.63 9.75 10.87 11.95 8.05 9.53 11.21
10.56 9.77 10.53 10.75 7.82 10.24 10.39
8.37 7.55 8.62
7.78 6.28 10.20
5.28 10.58 8.46
5.00 9.59 8.25
7.05 8.35 8.02 10.75 7.00 8.04 7.35
Table 8 provides average durations of unemployment spells by subsequent labour market states, that is the state that was moved into on leaving unemployment. For the sample as a whole (those entering the labour market since 1914), the longest unemployment spells were terminated by moving out of the labour force altogether (into education/training, retirement, long-term sick or other non-specified states). This suggests that perhaps exits from the labour market are caused by an inability to find suitable employment. More recent labour market entry cohorts show a slightly different pattern, with the durations of unemployment spells followed by selfemployment becoming longer. This may be an age effect as self-employment tends to be more popular among more mature members of the labour force; younger cohorts may be less willing to enter self-employment without first accumulating labour market experience. Evidence from the post 1990 sample supports this intuition, as
14
unemployment spells followed by self-employment are on average the shortest (at 5.7 months) for this group. As unemployment spells terminated by a move into education are on average consistently among the longest, it appears that individuals take education and training courses as a response to unemployment experiences. Table 8: Unemployment duration (months) by subsequent (destination) labour market state Subsequent labour market state Employment Self-employment Education/Training Retired Long term sick Other Total
Sample period 1914-
1946-
1973-
1980-
1990-
9.28 9.15 13.53 35.80 15.41 27.88 11.21
9.14 9.11 14.72 14.04 15.88 28.91 10.39
7.86 9.61 10.93
7.59 10.02 11.24
9.16 13.19 8.46
7.92 8.89 8.25
6.27 5.69 13.19 9.11 9.09 10.84 7.35
V. Summary and Conclusions This paper has established some stylised facts about the changing picture of male unemployment in Britain using retrospective life history and panel data from the BHPS. The data have revealed some important trends. More recent labour market entry cohorts have experienced more unemployment spells on average than earlier cohorts, and are more prone to repeated spells. These spells also tend to be shorter in duration, although individuals actually spend a greater proportion of their time unemployed. More recent cohorts are also less likely to experience redundancy, but suffer more dismissals and non-renewed temporary work8. As well as comparing labour market entry cohorts, the paper has also contrasted the experiences of the male sample as a whole with those experienced by men currently active in the labour market since 1990. This has shown that those currently active in the labour market are more likely to have moved between unemployment and self-employment than in previous years, and are less likely to have entered unemployment voluntarily. This is perhaps a reflection of the changes in the labour market that have occurred in the last fifteen years.
8
These findings are consistent with biases one might expect to arise from recall error, but disentangling the two effects remains problematical.
15
This has been a first step at investigating the kinds of issues that only retrospective and panel data can enlighten. In future we shall use the BHPS to investigate the characteristics of individuals experiencing repeated unemployment spells.
REFERENCES Campbell M and M Daly (1992), “Self-Employment: Into the 1990s”, Employment Gazette, June. Disney R (1979), "Recurrent Spells and the Concentration of Unemployment in Great Britain", Economic Journal, March, 109-119. Jackman R and Layard R, (1991) “Does Long Term Unemployment Reduce a Person’s Chance of a Job? A Time Series Test”, Economica, LVIII, 93-106. Hildreth AKG, SP Millard, D Mortensen and MP Taylor, (1995) “Wages, Work and Unemployment: Work History Evidence on Allocating Time Between Labour Market States”, mimeo, University of Essex. Nickell SJ, (1980) "A Picture of Male Unemployment in Britain", Economic Journal 90, 776-794. Taylor MF (ed), with J Brice, N Buck and E Prentice (1996) British Household Panel Survey User Manual Volumes A and B. Colchester: University of Essex. Taylor MP (1996) “Earnings, Independence or Unemployment: Why become SelfEmployed?”, Oxford Bulletin of Economics and Statistics, Vol 58, No 2. Van den Berg, GJ and Van Ours JC (1994), “Unemployment Dynamics and Duration Dependence in France, The Netherlands and the United Kingdom”, Economic Journal, CIV, 432-443.
16
APPENDIX
Table A1: Reasons given for leaving job immediately preceding an unemployment spell Reason 1914Different/better job Redundancy Dismissed Temporary job ended Retired Health reasons Family care Other Total number of spells
261 583 111 199 27 42 9 209
(18.1) (40.5) (7.7) (13.8) (1.9) (2.9) (0.1) (14.5)
1946253 529 104 187 10 41 8 195
1441
(19.1) (39.9) (7.8) (14.1) (0.1) (3.1) (0.1) (11.0)
Sample period 1973-
1980-
148 234 70 127
(20.9) (33.1) (9.9) (18.0)
72 129 54 96
(16.4) (29.3) (12.3) (21.8)
16 4 108
(2.3) (0.1) (15.3)
11 2 76
(2.5) (0.1) (17.3)
1327
707
199019 210 37 90 6 13 3 85
440
(4.1) (45.4) (8.0) (19.4) (1.3) (2.8) (0.6) (18.4)
463
Table A2: Voluntary or involuntary separation from previous employment Reason 1914Voluntary Involuntary Total number of spells
521 920
(36.2) (63.8)
1441
1946497 830
(37.4) (62.5)
1327
Sample period 1973276 431
(39.0) (61.0)
707
1980161 279
(36.6) (63.4)
440
1990113 350
(24.4) (75.6)
463
17
