Appendix 1:
New Estimates of Numbers in Low Pay

In order to make an accurate assessment of low pay in the economy, we need accurate data on the number of low-paid workers. While surveys of firms and qualitative research give us a feel for the number and characteristics of the low paid, only the official national data sources provide the aggregate picture.

Our second report set out, in some detail, the uncertainties surrounding official estimates of low pay. The Labour Force Survey (LFS) and the New Earnings Survey (NES) each produce biased estimates of the numbers in low pay for different reasons. The NES under-samples individuals earning less than PAYE thresholds, and therefore understates the level of low pay. The LFS suffers from problems of bias in its estimation of hourly earnings, which lead to an understatement of hourly earnings, and produce much higher estimates of numbers on low pay than the NES.

We asked the Office for National Statistics (ONS) to address these problems with the data and to improve the methodology for estimating numbers in low pay. The ONS has produced improved estimates, which were first released in a press notice on 26 October 2000, and in more detail in the January 2001 issue of Labour Market Trends. The revised methodology produced lower estimates of numbers on low pay than we presented in our second report. Here we discuss the reasons why the estimates have fallen, and what the new figures are saying.

Appendix 2 of the second report explained in some detail how the estimates of low pay were derived. Here we give a brief overview, as this background is necessary to understand why the estimates of numbers in low pay have changed.

• the LFS was adjusted to bring numbers with low wages down; and

• the NES was adjusted to include more people earning below PAYE levels.

The adjusted LFS figure produced the upper bound of the central estimate, and the adjusted NES figure gave the lower bound. In our first report, based on ONS advice, we used the mid-point between these upper and lower bounds to provide a central estimate of numbers in low pay. In our second report, again based on ONS advice, we argued that the NES estimate, the lower bound, was likely to be closer to the actual number of low paid than the central figure. This reduced the estimate of numbers in low pay to around 1.7 million.

The figure of 1.7 million was a combination of the NES data and the earnings data from the LFS. In Spring 1999, the NES suggested that around 10 per cent of employees earned less than the PAYE threshold, whereas the LFS estimated that around 17 per cent of cases did so. The methodology used in the second report assumed that the LFS gave the true percentage of people, and the true distribution of hourly earnings for those earning below PAYE. The number of low paid above PAYE was calculated using only NES data. The below and above PAYE parts of the distribution were added together to produce an overall ‘NES’ estimate of low pay.

At the time of writing the second report, new work by ONS exploring the accuracy of the NES data suggested that the bias in the NES was not as great as was first thought. The second report therefore predicted that estimates of numbers in low pay might fall below the lower bound estimate of around 1.7 million, more in the region of 1.5 to 1.7 million.

The full methodology underlying current estimates was set out in the January 2001 issue of Labour Market Trends. The new methodology provides two separate estimates: one based solely on NES data, the other solely on LFS data. The central estimate is the mid-point between the two estimates.

The LFS estimate is now based on the results of a new question, introduced into the LFS in April 1999, that asks individuals their actual hourly rate of earnings. (The previous method for deriving hourly wage rates was to divide gross weekly earnings by usual weekly hours.) This has improved the reliability of the LFS as a means of estimating hourly pay, as estimates of hourly rates are no longer derived from inconsistent earnings and hours information.

• For post-1999 data a donor method of imputation is used. This involves using information on an individual’s earnings and hours worked, and such personal characteristics as age, sex and qualifications, to predict a value for their hourly rate. The donor method, which is valid only for the post-minimum wage period, then groups individuals with an actual value of the hourly rate, the ‘donor’, with other individuals with similar predicted hourly rates. Individuals without an actual hourly rate are then randomly assigned an actual hourly rate from a similar donor. Similar donors are found through a regression to find hourly rates. The main reason for using donors to get the final imputed values is because there is a large spike in the earnings distribution at the level of the National Minimum Wage. The donor method replicates that spike.
• The donor method is not used to estimate earnings for the period before April 1999. There is no spike in the earnings distribution pre-minimum wage, and using the donor method would have reproduced the 1999 spike in the 1998 distribution. The reason for this is that donors would have had to be drawn from the 1999 data set because the 1998 data do not include the hourly rate variable. Hourly rates for 1998 are therefore estimated using predictions from the regression itself.

The NES estimate was produced by adjusting the NES so that it better reflected the characteristics of all employees and the firms in which they worked. Comparing the NES with other data sources suggested that it under-represented the number of young people and people working in small firms. This under-representation was adjusted, although not completely corrected, by grossing up the NES sample to all employee jobs in the UK. The grossing regime ensured that there was the right number of jobs in each cell of the grossing matrix, but if any given cell was not representative of the whole population, grossing could not correct for this. For example, if the NES under-represented young people in the hotel and catering industry, grossing would ensure the right number of this group, but would not compensate for the lack of low earners among them in the original sample. Hence the grossed NES data are still likely to underestimate the numbers earning below National Minimum Wage rates.

The population matrix used for grossing developed by the ONS had control totals for individual characteristics (age group and gender) and firm characteristics (number of employees and sector). The individual level data came from the LFS, while the firm and sector level data came from the Inter-Departmental Business Register (IDBR). The IDBR contains information on all businesses that are registered for VAT or PAYE purposes. Grossing factors were produced for each cell in the matrix.

The central estimate, derived from both the NES and the LFS data, is an average of the proportion below the National Minimum Wage estimated from each of the two data sources. In order to estimate the number below the minimum wage, this proportion was applied to the total number of jobs in the economy from the LFS.

In the second report we explained that, in order to estimate the number of potential beneficiaries from the National Minimum Wage, we needed to downrate the level of the minimum wage to represent its real value in 1998. We therefore downrated the level of the National Minimum Wage to £3.50/£2.90 in 1998 to allow for increases in earnings over the period. The estimate of the number of jobs below £3.50/£2.90 in 1998 based on the new methodology is around 1.3 million.

• it does not have the very low values seen in the derived variable;
• it is more up-to-date;
• it measures only basic pay;
• it is sensitive to the introduction of the National Minimum Wage, whereas the old variable showed no discernible response to the introduction of the minimum wage; and
• it has a high level of personal response for those aged 22 and over, and evidence suggests that proxy respondents do not guess if they do not know.

Nonetheless, the LFS is a survey of households and it will be subject to inaccuracies and response error. The method of imputing hourly wages for people who do not answer the hourly rate question will increase the level of uncertainty about the actual hourly rate of people in the sample. The imputed answers will be estimates based on the old and imperfectly derived hourly rate variable.

A further reason for caution is that the stability of the LFS estimate over time has not been fully tested. Since Spring 2000 questions concerning the new variable have been asked of a greater proportion of the LFS sample, this may lead to a further refinement of the estimate of low pay. Also, as explained above, the difference between the regression-only and the donor method results suggests that the 1998 estimate of low pay using the regression-only method may have led to an overestimate on the part of the LFS.

Since it is derived from pay records, the NES gives more accurate estimates of hourly wages for those people in the sample. The nature of the sample, however, is to under-represent the number of people in low pay. The grossing of the NES is also an improvement on the old methodology. Comparing the grossed and ungrossed data shows that the grossed NES more fully reflects the composition of all employees. Tables A1.2 to A1.4 below give comparisons of the ungrossed NES with estimates of the number of employees by size of business and sector, and estimates of number of employees by age and sex from the data sources used in the grossing procedure.

In summary, we believe that the new methodology is better than the previous methods. But we remain concerned that there is still considerable uncertainty surrounding both estimates of low pay. This has a direct impact on our work and our ability to have confidence in the data when we consider our recommendations. We would therefore not want to rely too heavily on either one of these estimates alone, but to consider them together. The nature of the estimates and their weaknesses must be borne in mind when considering the low-pay estimates.

We were encouraged by the convergence of the estimates of low pay from the two data sources at low hourly wage rates. The difference between the NES and LFS estimates is much closer than the upper and lower bounds used for our second report. The bottom of the range is around 1.2 million (NES) and the top of the range 1.4 million (LFS). The closeness of the two figures tends to suggest that the central estimate is a better estimate of the numbers on low pay than we have had hitherto. Estimates of low-paid workers are a vital tool to inform the work of the Commission. They assist us to formulate our recommendations. We are grateful to the ONS for the work it has done in improving the data. We hope that it will continue to do further work on producing better estimates.

Figure A1.2

Cumlative Hourly Earnings Distribution, 1998

Source: ONS, 1998

Since we first embarked on our task of advising the Government on the National Minimum Wage, we have had concerns about the estimates of those in low pay. We wrote about these in our first and second reports. We are grateful to the ONS for the positive way in which it has reacted to our concerns and the work that it has done to improve the statistics on low pay. This appendix gives our view on the new methodology ONS has developed and which has produced a new central estimate. Although it provides a more robust estimate of the extent of low pay in the economy, uncertainties remain. We are particularly concerned about the estimate derived from the LFS, which could in future add downward pressure to the central estimate.

In our first report, we estimated that the rate of the National Minimum Wage we had recommended would benefit around 1.9 million people, or about 9 per cent of those in work. Following revisions to the data last year, this number fell to 1.5 million. The latest estimates put this number at 1.3 million. We urge the ONS to continue to work on improving the estimate further so that we can be better informed in advising the Government, and so that the Government can be better informed on the extent to which the National Minimum Wage is benefiting the low paid.