Recent shifts in wage inequality and the wage returns to education in Britain.

Machin, Stephen

1. Introduction

Since the late 1970s wage inequality has risen at a rapid rate in Britain, and the 1980s were characterised by the rich substantially improving their labour market position, with the poor falling sharply behind. But what has happened since then? Has the trend continued? And, if so, is it in the same form seen in the 1980s, with big demand shifts in favour of relatively more skilled workers?

In this article I use several individual- and industry-level data sources to examine these questions. Each source paints a similar picture, namely one of more slowly rising wage inequality in the 1990s. There has been a clear deceleration in the rate of change of the still widening gap between the rich and poor. Unless there is a remarkable acceleration in the last few years of this decade, the 1980s will clearly go down as the decade of the 20th century where wage inequality rose by most.

The 1990s have also seen rapid educational upgrading which has translated into big increases in the relative supply of more educated workers. The late 1980s reforms to the education system (notably the introduction of the GCSE system of examinations) have compounded the expansion of the higher education system, and generated an acceleration of the relative supply of highly educated labour. This acceleration in supply is probably what lies behind the slowdown in the rate of increase of wage inequality in the 1990s.

But as wage gaps are probably still rising, albeit at a reduced rate, it seems that they have not been tempered enough by rising supply to meet employers' demand requirements. As the wage returns to education have still risen in the face of increased supply, this must mean that the 1990s, like the 1980s, was characterised by relative demand shifts in favour of the skilled. In line with some of my earlier work (Berman, Bound and Machin, 1998; Machin, 1996a, 1996b; Machin and Van Reenen, 1998), I see nothing to change the view that technological changes that have altered the nature of work, and the way it is rewarded, are central to the changes in labour market inequality observed in recent decades.

The structure of the article is as follows. Section 2 reports on what has happened to indices of wage inequality in the 1990s, being careful to contrast them with the 1970s where wage inequality fell, and the 1980s where it rose very sharply. Section 3 reports on shifts in the wage returns to education, on the supply changes that have accompanied them and on the relative demand changes they together imply. Section 4 considers how these relative demand shifts are related to observable indicators of technology and Section 5 then concludes.

2. Changes in wage inequality in the 1970s, 1980s and 1990s

Data sources

There are several sources of data for looking at wage inequality trends in Britain, all of which have their relative merits and shortcomings. The longest time series are available from the household-level Family Expenditure Survey (FES) and the employer reported individual-level New Earnings Survey (NES). The potential drawbacks with the FES are its relatively small sample size, as it covers only around 10,000 people in work each year, and the fact that it does not contain any data on educational qualifications.(1) Sample size is not a problem for the NES which, in principle, covers 1 per cent of the working population (those whose National Insurance numbers end in 14), thereby generating large sample sizes (in what follows these are of the order of over 100,000 per year). But it should be noted that the NES has its own problems if one wishes to consider inequality trends: (i) it under-samples low wage part-time workers; and (ii) there are no data on education.

Nevertheless for most of what I consider in terms of trends in wage inequality I focus on these two sources, due to their long time series and because they both report data on hourly wages. The other data source I draw upon, the General Household Survey (GHS), does not report hourly wages on a consistent basis through time, though it too spans a long time period and also has the big advantage of reporting data on the highest educational qualification of individuals. So, when I turn to shifts in relative wages across education groups in Section 3 I use mainly the GHS.(2)

Trends in Gini coefficients

Table 1a reports Gini coefficients for male earnings measures based on the GHS, NES and FES between 1975 and 1996. Table 1b reports the same for female earnings. All are based on earnings expressed in 1996 prices. The pattern the Gini coefficients imply is clear, irrespective of whether one looks at weekly or hourly earnings, and is rather similar by gender. There was a slight fall in inequality in the 1970s, up to around 1977 or 1978 (depending on which data source one looks at), after which inequality rises. The scale of the 1970s fall (which commentators at the time thought corresponded to a large compression) clearly pales into insignificance compared with the rises that followed.

From the late 1970s and through the 1980s the inequality of earnings rose massively for both sexes. For example, according to the FES Gini coefficients for male hourly earnings, the rise was from 0.240 in 1979 up to 0.312 by 1990, corresponding to a 30 per cent increase over 11 years. For females, the Gini rose from 0.252 to 0.320, corresponding to a 27 per cent increase.

Table 1a. Gini coefficients, male wages 1975-96 (wages in 1996
prices)

Men Weekly wages Hourly wages
Year GHS NES FES NES FES

1975 0.249 0.236 0.244 0.223 0.239
1976 0.242 0.239 0.240 0.229 0.235
1977 0.240 0.233 0.238 0.221 0.236
1978 0.237 0.239 0.244 0.227 0.244
1979 0.231 0.245 0.246 0.228 0.240
1980 0.244 0.245 0.262 0.233 0.256
1981 0.250 0.251 0.266 0.246 0.264
1982 0.277 0.254 0.260 0.248 0.258
1983 0.256 0.260 0.279 0.253 0.276
1984 0.281 0.274 0.276 0.261 0.273
1985 0.283 0.275 0.292 0.261 0.285
1986 0.294 0.279 0.298 0.266 0.287
1987 0.301 0.289 0.319 0.277 0.305
1988 0.298 0.295 0.311 0.284 0.299
1989 0.302 0.298 0.302 0.288 0.291
1990 0.301 0.299 0.324 0.289 0.312
1991 0.313 0.303 0.321 0.294 0.301
1992 0.316 0.306 0.330 0.296 0.309
1993 0.314 0.311 0.338 0.301 0.319
1994 0.324 0.316 0.338 0.308 0.320
1995 0.325 0.322 0.326 0.317 0.358
1996 - 0.325 0.329 0.318 0.335
Table 1b. Gini coefficients, female wages 1975-96 (wages in 1996
prices)

Women Weekly wages Hourly wages
Year GHS NES FES NES FES

1975 0.349 0.281 0.356 0.215 0.256
1976 0.350 0.294 0.360 0.225 0.261
1977 0.349 0.288 0.353 0.206 0.244
1978 0.347 0.290 0.353 0.205 0.258
1979 0.342 0.295 0.355 0.203 0.252
1980 0.350 0.300 0.359 0.210 0.253
1981 0.365 0.313 0.377 0.231 0.279
1982 0.371 0.315 0.381 0.227 0.270
1983 0.364 0.323 0.388 0.230 0.281
1984 0.375 0.337 0.381 0.236 0.277
1985 0.390 0.335 0.387 0.236 0.283
1986 0.390 0.340 0.386 0.237 0.298
1987 0.400 0.348 0.397 0.240 0.311
1988 0.392 0.355 0.394 0.250 0.310
1989 0.394 0.362 0.393 0.260 0.295
1990 0.395 0.366 0.401 0.260 0.320
1991 0.401 0.370 0.396 0.268 0.300
1992 0.402 0.377 0.409 0.270 0.306
1993 0.414 0.377 0.403 0.277 0.306
1994 0.403 0.380 0.403 0.278 0.307
1995 0.404 0.386 0.407 0.289 0.300
1996 - 0.388 0.400 0.288 0.312

[TABULAR DATA FOR TABLE 2A OMITTED]

[TABULAR DATA FOR TABLE 2B OMITTED]

But what of the 1990s? The Gini coefficients seem to show inequality continuing to rise, but at a slower rate. For example, the FES Gini for male hourly wages rises from 0.312 in 1990 to 0.335 by 1996, corresponding to a rise of just over 7 per cent, or 1.2 per cent a year. But the Gini rose by over twice as much a year in the 1980s at 2.7 per cent a year. For women the slowdown is even more marked. Between 1990 and 1996 the FES Gini for hourly wages actually fell, from 0.320 to 0.312. Looking at the other data sources shows similar patterns in terms of comparisons between the 1980s and 1990s. In all cases the 1979-90 annualised change in the Gini is higher, usually considerably higher, than the 1990-96 annualised change.

Changes at different percentiles of the distribution

The Gini coefficient is a single number that describes the inequality of the earnings distribution at a point in time and as such is useful for considering aggregate inequality trends. It is, however, less useful for identifying shifts that lie behind the overall change. So Table 2a reports on what has happened to male real hourly earnings at different points in the distribution by reporting the level of earnings at each decile cutoff of the distribution from the Family Expenditure Survey. It also reports 90-10, 50-10 and 90-50 ratios in the final three columns. Table 2b reports equivalent FES numbers for women.

First consider the 90-10 wage ratio. Reassuringly this shows much the same pattern as the Gini's in Tables 1a and 1b. Wage inequality for both men and women rises very sharply from the late 1970s to the late 1980s/early 1990s, but very much slows down in the 1990s. The ratios describing the evolution of the lower end of the distribution (the 50-10 ratio) and the upper end (the 90-50 ratio) demonstrate that the 1980s rise in wage inequality was characterised by an opening out at both ends of the distribution with the highest earners doing much better than those in the middle, but in turn the middle doing much better than the bottom. But in the 1990s wage inequality stopped rising.

In terms of growth at particular decile cutoffs Chart 1 shows the annualised per cent change in hourly wages at the 10th, 50th and 90th percentiles in the 1970s, 1980s and 1990s. The 1980s rise in wage inequality is illustrated very clearly as wage growth at the 90th percentile outstrips the 50th and 10th by some distance. But the 1990s is very different, showing low real wage growth at all percentiles, especially for men where real wage growth is barely above zero. This gender difference in the 1990s, where women seem to be doing better than men at all points of the distribution, is something I will return to later.

The numbers reported so far cover the wage distribution for all workers. One may also be interested if there are any differences if one focusses only on full-time workers, particularly given the compositional changes in the nature of employment that happened over the time period I consider (especially relevant is the rise in part-time female employment). Chart 2 therefore presents the annualised per cent change in hourly wages at the 10th, 50th and 90th percentiles in the 1970s, 1980s and 1990s for full-timers. Not surprisingly the only differences that emerge are at the lower end of the distribution. Limiting to full-timers shows a bigger compression in the 1970s, due to higher wage growth at the 10th percentile, than that revealed by looking at all workers.(3) Nevertheless the overall pattern remains, with wage inequality falling a little in the 1970s, followed by the rapid rise of the 1980s, and then a slowdown in the 1990s.

3. Changes in relative supply and demand

It has been noted in many places that the 1980s rise in wage inequality displays an apparent contradiction with simple supply-demand models in that there were large wage gains for workers at the upper end of the wage distribution despite there being many more workers with higher levels of human capital.(4) The most commonly cited example of this is the fact that the supply of more highly educated workers rose in the 1980s but, at the same time, the wage returns to higher educational qualifications also went up. The way this is usually rationalised is by stating that employers' demand for more highly educated workers must have increased by more than the supply, thereby explaining why their relative wages rose.

In this section I consider these mechanisms that lie behind changes in wage inequality in more detail. I first focus on what has happened to the relative supply of workers with different education levels over time, and then look at what happened to their relative wages. I then consider the implications of these changes for explaining the evolution of wage inequality in the 1970s, 1980s and 1990s.

Changes in relative supply

Educational upgrading has been occurring continuously for many years. Far more recent entrants to the labour market are now graduates than before and far fewer individuals now leave school with no formal educational qualifications. For example, in 1975 around 6 per cent of employed men and 2 per cent of employed women had a degree or higher educational qualification. On the other hand 50 per cent of employed men and 58 per cent of employed women had no educational qualifications. By 1994 15 per cent of men (10 per cent of women) in the workforce had a degree, and only 22 per cent of men (24 per cent of women) had no educational qualifications.(5)

This corresponds to substantial educational upgrading or, put another way, large relative supply shifts since there are now far more highly educated workers entering the labour force. In this article I consider the extent of supply changes for two comparison groups, the number of workers with a degree versus the number of workers without a degree and the number of workers with a degree versus the number without any educational qualifications. For comparisons based on two groups of workers at a point in time the relative supply of group 1 workers as compared to group 2 workers is simply the ratio of their employment levels [L.sub.1]/[L.sub.2] so between two time periods (say s and t) one can define changes in relative supply between the two groups as [Delta][S.sub.12] = [log[([L.sub.1]/[L.sub.2]).sub.t] - log[([L.sub.1]/[L.sub.2]).sub.s]]/(t - s).

Table 3 reports annualised changes in relative supply ([Delta][S.sub.12] expressed as percentage changes) in Britain in the 1970s, 1980s and 1990s based on General Household Survey data for the degree/no-degree and degree/no qualifications comparisons. The pattern is very clear and points to supply shifts towards the more educated groups in all time periods. The time series pattern is however of some interest, with supply rising rapidly in the 1970s, slowing down in the 1980s and accelerating again in the 1990s. This is true for both comparisons and for men and women.

Without any change in employers' demand patterns one would normally think that these increases in relative supply should dampen down the wages of the more educated as compared with the less educated as there are more of them for employers to choose from. But this has not happened as the comparison of changes in the relative wages of these groups that I next consider demonstrates.

Change in relative wages

Table 3 also reports the change in relative wages for the two comparison groups of interest. The wage changes are derived from estimated coefficients on education dummy variables included in semi-log earnings equations estimated for full-time employees in different time periods. For person i in period t the wage equations take the following form

log([W.sub.it]) = [[Alpha].sub.t] + [[Beta].sub.t][D.sub.it] + [[Gamma].sub.t][X.sub.it] + [[Epsilon].sub.it] (1)

where log(W) is log earnings, D is a dummy variable distinguishing between the education groups of interest, X is a set of control variables(6) and [Epsilon] is an error term.

Table 3 reports annualised log changes in education based wage differentials over the 1970s, 1980s and 1990s. For two particular time periods s and t these correspond to [Delta][[Omega].sub.12] = {[[[Beta].sub.t] - [[Beta].sub.]]/[t - s]}.sup.*] 100. Two sets of changes are reported in each case, one from the equation (1) specification, the other which additionally includes a Heckman (1979) selectivity correction term in the earnings equation.(7)

Changes in the estimated wage differentials show a familiar pattern, with falls in the 1970s and rises in the 1980s. Interestingly in almost all cases the wage differentials also seem to go up in the 1990s. In terms of the [TABULAR DATA FOR TABLE 3 OMITTED] nature of relative supply changes over time the pattern is in line with the idea that relative wages alter in response to supply. However, given the magnitude of the supply changes and the lack of relative wage falls, there must be something else going on, namely that demand must have also shifted in favour of more educated workers. I consider this in the next sub-section.

Shifts in relative demand

Whilst the 1970s sees rising relative supply coupled with falling relative wages this is not true of the periods from 1980 onwards as there is a positive covariation between changes in relative supply and changes in relative wages. This seems to suggest that relative demand shifts in favour of the more highly educated must have happened.

One can define a measure of the implied relative demand shifts between two sort of workers as [Delta][D.sub.12] = [Delta][S.sub.12] + [Sigma][Delta][[Omega].sub.12], where [Sigma] is the elasticity of substitution between group 1 and 2 type workers. Then, given an estimate of [Sigma], [Delta][D.sub.12] can be calculated for our comparison groups of interest in the 1970s, 1980s and 1990s.(8) But before doing this note that [Sigma] measures how easily one can substitute group 1 and 2 workers and is an index lying between 0 (no substitution) and [infinity] (perfect substitutability). So, according to the definition of [Delta][D.sub.12] the 1980s and 1990s must have seen demand shifts in favour of more educated workers as both [Delta][S.sub.12] and [Delta][[Omega].sub.12] are positive. In the 1970s it is mathematically possible that there was no relative demand shift but this requires a large estimate of [Sigma]: for the demand shift to equal zero [Sigma] would need to be 6.54, which is considerably higher than the estimates in the literature (see Hamermesh, 1993).

Table 3 reports estimates of the implied relative demand shifts in favour of the more highly educated under the assumption made by Katz and Murphy (1992) and Autor, Katz and Krueger (1997) that [Sigma] = 1.4. It should be noted that adopting this assumption is not critical unless [Sigma] lies much outside the 1 to 2 range identified by most empirical studies. The other point to note is that, in the absence of any evidence on the issue, I assume the same value in all sub-periods. Again, unless [Sigma] has changed substantially over time (which I doubt), the results will not be much affected by this.

The values of [Delta][D.sub.12] given in the table show relative demand shifts in favour of the more educated in all three sub-periods. Put alternatively, demand must have [TABULAR DATA FOR TABLE 4 OMITTED] moved towards the more educated as their relative wages simply did not respond enough to the large increases in relative supply that have occurred. In terms of the degree/non-degree comparisons these relative demand shifts seem to have occurred in each sub-period and seem to be continuing into the 1990s. This pattern is the case for both men and women, although the demand shifts are larger in magnitude for women. For the degree/no qualifications comparison demand has shifted very markedly against the no qualifications group, despite there being far fewer of them, sharply illustrating the rapidly deteriorating labour market position of unskilled workers.(9) Again the demand shifts are larger for women than men, reflecting the fact that gender based earnings differentials have fallen over this time period (Harkness, 1996), despite faster increases in relative supply for women as compared to men.(10)

The evidence presented in this section clearly shows that, whilst changes in relative supply seem to have contributed to the 1990s slowdown in rising wage inequality, demand has continued to shift in favour of more educated and skilled workers. I show in the next section that these shifts in demand are inherently connected to the technological progressiveness of workplaces and firms which increasingly demand more skilled workers.

4. Shifts in relative demand and technology

There has been much discussion about what lies behind the relative demand shifts in favour of more educated and skilled workers that seem to have occurred in many industrialised countries. Much of this has focussed on the relative merits of explanations that emphasise technology, in particular the introduction of skill-biased technological changes, and rises in international competition, particularly with low wage countries from the southern hemisphere. Some of my earlier work emphasises the links with technology (Machin, 1996b; Berman, Bound and Machin, 1998; Machin and Van Reenen, 1998) and the same work finds it hard to identify any links with observable measures of changing international trade.(11)

Table 4 summarises some of these findings and also reports some results that extend into the 1990s. The table reports coefficient estimates on technology measures included in industry-level skill upgrading equations, where skill upgrading is measured by changes in the skilled wage bill share.(12) Two measures of skill are adopted, the nonproduction wage bill share in columns (1)-(3) and the graduate wage bill share in columns (4) and (5).

[TABULAR DATA FOR TABLE 5 OMITTED]

Columns (1) and (2) are taken from Machin (1996b) and show that more R&D intensive and more innovative industries were those that experienced faster increases in non-production wage bill shares in the 1980s. Column (3) shows the same to be true of industries with higher computer usage. Perhaps more interesting here are columns (4) and (5) which present regressions of changes in the graduate wage bill share on computer usage from GHS data which extends into the 1990s. In both cases (column (4) for men and column (5) for women) the graduate wage bill share rose by more in industries with higher computer usage. The similarity of this industry-based evidence that continues into the 1990s is in line with the idea that technology is still shaping the more recent changes in skill structure that have occurred.

The fact that technology must be important is also revealed by cross-country evidence. When one considers identically specified regressions of industry skill upgrading on R&D intensity for the same industries in different countries one always seems to uncover a positive association. Table 5 reports coefficients on R&D intensity (R&D/Value added) for regressions in six countries (Denmark, Germany, Japan, Sweden, the UK and US) and in all cases a positive (significant) association is uncovered. Such regularities are uncommon and not easy to find in international comparisons and suggest an important link between the extent of demand shifts in favour of the more skilled and observable measures of technology.

Overall, the connections between changes in relative demand and technology seem to be present in terms of comparisons of different measures of skill upgrading and technology, across different time periods and across countries. Robust evidence like this based on direct measurable indicators is not forthcoming for other possible explanations of why the skilled have done much better (and the unskilled much worse) in recent years and hence I see little reason to conclude in any other way than by reiterating the point that an important part of the rise in labour market inequality is inherently linked to technology.

5. Conclusions

This article has considered overall shifts in wage inequality in Britain between the mid-1970s and mid-1990s, showing that the key decade of rising wage inequality in Britain was the 1980s. In the 1990s the rising gap between the highest and lowest paid was either stable or rose a little, but by nowhere near as much as in the 1980s. This seems to be, at least partially, due to the fact that faster educational upgrading has dampened down some of the rising wage differentials experienced by the more educated. Nevertheless, demand still seems to be shifting in favour of the more highly educated and skilled. Put another way, despite the fact that there are many more workers with higher educational qualifications, employers' demand for them is still rising, hence their wages relative to other groups have not fallen.

So, the patterns revealed by considering recent shifts in wage inequality in Britain do not seem to suggest anything very new happening in terms of the relative labour market position of more and less educated workers. Whilst wage inequality has not risen as fast as in the 1980s, rising supply has still translated into relative demand shifts in favour of the more educated as the wages of the more educated have not fallen. Such demand shifts are still more pronounced in more technologically advanced industries which is in line with the notion, like much of the evidence based on industry demand shifts in the 1970s and 1980s, that technology is key to changes in labour market inequality.

Comments should be addressed to the author at the Department of Economics, University College London and Centre for Economic Performance, London School of Economics. This work draws on some DfEE funded work (joint with Susan Harkness) on 'Graduate Earnings in Britain, 1974-95' and on some of my joint work with John Van Reenen (Machin and Van Reenen, 1998). I would also link to thank Richard Dickens, Paul Gregg and Julian Steer for their help with some of the data.

NOTES

(1) It does contain information on years of schooling since 1978.

(2) The other large scale microdata source I could use is the Labour Force Survey (LFS) which, like the GHS, has good education data going back to the 1970s. Unfortunately wage data have only been collected since 1992 and, as I am interested in longer-run changes in wage inequality, this is not much use here.

(3) See Harkness (1996) for more details on differences in the evolution over time of the full-time and part-time female earnings distributions.

(4) See Bound and Johnson (1992) and Katz and Murphy (1992) for early expressions of this in the US, and Johnson's (1997) survey piece. In the UK context see the discussion in Machin (1996a) and Schmitt (1995).

(5) These numbers are taken from the General Household Survey data used in Harkness and Machin (1998).

(6) The control variables entered in the earnings equations were age, age squared, industry and region dummy variables and a dummy variable for whether the individual is a teacher.

(7) The selection term is identified by including non-labour income and a dummy variable for whether another adult in the household is in work in the first stage employment model and by excluding these variables from the earnings equation. As it turns out, whilst comparisons of the cross-section estimates of [Beta] are sometimes affected by the inclusion of the selection term, the overall pattern of change is, for the most part, robust to the selection corrections (see Harkness and Machin, 1998).

(8) See Autor, Katz and Krueger (1997) for calculations of this sort based on US Census and Current Population Survey data.

(9) One can plausibly argue that workers with degrees are really not very substitutable with workers with no educational qualifications (i.e. should be lower for the degree/no qualifications comparison). Even in the extreme case of [Sigma] = 0 this would still imply larger shifts in the demand for graduates vis-a-vis workers with no qualifications as compared to the graduate/non-graduate comparisons.

(10) Focussing on recent labour market entrants where the supply changes are even faster and accelerate sharply in the 1990s does show smaller 1990s wage changes but still produces sizeable implied relative demand shifts in favour of the more educated (see Harkness and Machin, 1998). This is in line with the work emphasising higher wage inequality for more recent cohorts entering the labour market (see Gosling, Machin and Meghir, 1995).

(11) See also Desjonqueres, Machin and Van Reenen (1998) which identifies relative demand shifts in favour of the skilled in non-traded industries that have not been exposed to international trade.

(12) Recall from the discussion above that, for an elasticity of substitution near unity, the skilled wage bill share measures the relative demand shift between skilled and unskilled workers.

REFERENCES

Autor, D., Katz, L. and Krueger, A. (1997), 'Computing inequality: have computers changed the labor market', NBER Working Paper 5956.

Berman, E., Bound, J. and Machin, S. (1998), 'Implications of skill biased technological change: international evidence', forthcoming in Quarterly Journal of Economics.

Bound, J. and Johnson, G. (1992), 'Changes in the structure of wages in the 1980s: an evaluation of alternative explanations', American Economic Review, 92, pp. 371-392.

Desjonqueres, T., Machin, S. and Van Reenen, J. (1998), 'Another nail in the coffin? Or can the trade based explanation of changing skill structures be resurrected?', mimeo.

Gosling, A., Machin, S. and Meghir, C. (1995), 'The changing distribution of male wages in the UK', Centre for Economic Performance, LSE, Discussion Paper 271.

Hamermesh, D. (1993), Labor Demand, Princeton University Press.

Harkness, S. (1996). 'The gender earnings gap: evidence from the UK', Fiscal Studies, 17, pp. 1-36.

----- and Machin, S. (1998), 'Changes in the wage returns to education in Britain: supply changes and the evolution of wage differentials by sex and subject of study', mimeo.

Heckman, J. (1979), 'Sample selection bias as a specification error', Econometrica, 47, pp. 153-61.

Johnson, G. (1997), 'Changes in earnings inequality: the role of demand shifts', Journal of Economic Perspectives, 11 (Spring), pp. 41-54.

Katz, L. and Murphy, K. M. (1992), 'Changes in relative wages, 1963-1987: supply and demand factors', Quarterly Journal of Economics, 108, pp. 35-78.

Machin, S. (1996a), 'Wage inequality in the UK', Oxford Review of Economic Policy, 7, pp. 49-62.

----- (1996b), 'Changes in the relative demand for skills in the UK labor market', in Booth, A. and Snower, D. (eds.), Acquiring Skills: Market Failures, Their Symptoms and Policy Responses, Cambridge: Cambridge University Press.

----- and Van Reenen, J. (1998), 'Technology and changes in skill structure: evidence from seven countries', forthcoming in Quarterly Journal of Economics.

Schmitt, J. (1995), 'The changing structure of male earnings in Britain, 1974-88', in Freeman, R. and Katz, L. (eds.), Changes and Differences in Wage Structures, University of Chicago Press.