RDP 2000-08: Nominal Wage Rigidity in Australia 3. Interpreting Nominal Wage Changes

3.1 Overseas Studies

There is a burgeoning empirical literature on nominal wage behaviour in the United States. Despite the array of studies, however, there is no consensus about the extent of wage rigidity. This lack of consensus stems from debate about the information content of the data sources that are used. Some US studies utilise firm-level data collected from payrolls and find a variety of outcomes, with Wilson (1999) and Altonji and Devereux (1999) finding strong evidence that wage changes are truncated at zero while others, such as Blinder and Choi (1990), find a surprising frequency of wage cuts. The inconsistent results of firm-level studies suggest that firm-specific shocks and their timing may be important, so that the results should not be generalised.

Nationally representative studies are more prominent in the literature. A substantial number draw on household information from individuals interviewed in the US Panel Study of Income Dynamics (PSID). They tend to find a clear asymmetry in wage-change distributions with bunching at zero change, but still identify a non-trivial share of workers who receive nominal pay cuts. However, at issue is whether or not the observed falls in pay are indicative of reporting error.^[9] Authors have constructed various hypothetical wage-change distributions based on alternative treatments of reporting error, and drawn quite different conclusions about the actual wage-change distribution in the United States (McLaughlin 1994; Akerlof et al 1996; Kahn 1997).^[10]

A recent innovation has been to identify wage-change distributions from the US employment cost index. This provides nationally representative information but, because it draws on employers' records, avoids the reporting errors associated with surveys of householders. Using this approach, Lebow, Saks and Wilson (1999) find stronger evidence of skewness away from wage cuts than those studies based on the PSID or other household surveys.

Similar exercises contrasting evidence from an employment cost index with household survey data have been conducted for New Zealand, where there was particular interest in the degree of wage flexibility following the adoption of an inflation target and the introduction of the Employment Contracts Act 1991 (Chapple 1996; Cassino 1995). While some non-trivial nominal wage cuts have been identified, most studies report a clear asymmetry of wage changes with an over-representation of zero wage changes. Again, wage data from an employment cost index tend to display greater skewness away from nominal wage cuts than data taken from household surveys.

The disparate results of the US and New Zealand studies highlight the sensitivity of estimates of the dispersion of wage changes to data sources and methods and invite a review of the measurement issues that accompany the choice of alternative sources of Australian wage data.

3.2 Choosing Suitable Wage Data

An ideal measure of wage rates would be the hourly cost of employing constant-quality labour to perform a given job. Most published measures of wages, however, differ from this ideal. Many standard series of earnings are collected as a level of earnings per person.^[11] For a variety of reasons, these measures of average earnings can change, even if hourly wage rates remain constant, and can give false signals about wage flexibility.

For example, a change in the average hours of paid work will change earnings per person, even if hourly wage rates remain constant. Similarly, if different classes of employees have different levels of earnings, changes in the composition of the workforce will change average earnings independently of changes in wage rates. Furthermore, sampling problems may arise, where new respondents enter the wage survey, but have different characteristics and wage levels to those that have exited, inducing volatility into measured wages growth that does not reflect changes in wage rates. Finally, over time, the workforce tends to progress, or ‘drift’, to higher-paid jobs, or enjoy salary increments in existing jobs, so that earnings per person tend to rise even if entry-level wage rates for given jobs remain unchanged.

Each of these effects makes earnings per person a biased measure of wages growth. When present, they distort the mean rate of measured wages growth, but may also distort the distribution of wage changes, obscuring the true nature and extent of wage rigidity. Increasingly, the implications of these measurement issues are gaining attention in the academic literature (Abraham, Spletzer and Stewart 1999; Krueger 1999).

In Australia, the problems associated with measures of earnings per person motivated the development of the Statistician's new Wage Cost Index (WCI), which aims to measure changes in wage rates for a precisely defined fixed basket of jobs, controlling for the quantity and quality of work done.^[12] Consequently, the index is unaffected by changes in hours worked, compositional changes in employment or wages drift (although it may be affected by sampling problems). In fact, the focus on pricing labour to constant quality brings the WCI closer to the analytical concept of a wage rate than the employment cost indices used in most other overseas studies of wage behaviour.^[13]

The WCI is, therefore, the preferred series for the examination of nominal wage behaviour. However, a detailed distribution of wage changes underlying the series is not publicly available. Furthermore, the series commences in December 1997, precluding examination of how the relationship between changes in inflation and nominal wage rigidity has evolved.^[14] This limited time series is disappointing, since identifying how wage-change distributions respond to changes in inflation is central to identifying whether the nominal rigidity is more binding, and therefore costly, when inflation is low.

The approach taken in this paper is to use a data set from Mercer Cullen Egan Dell (MCED) that, while inferior to the WCI, still has a number of desirable properties. Furthermore, it is available in a time series that spans periods of high and low inflation.

3.3 The Mercer Cullen Egan Dell Survey

For more than 30 years, Mercer Cullen Egan Dell has conducted a series of regular surveys of remuneration. The main purpose of these surveys is to provide clients with information about prevailing market rates of pay for specific job descriptions. Reflecting this, MCED conducts detailed job-matching exercises across participating firms in each survey to identify over 450 different positions for which remuneration can be compared. (Positions for which award-only rates apply are excluded from the sample.) Drawing on payroll information, firms then report the remuneration for each employee in that position. Remuneration is broken down into base pay and total pay. Since the mid 1980s, pay for those who have remained employed in a given position between surveys (same incumbents or ‘stayers’) is separately identified from pay for all individuals in that position. The mean and the quartiles of the distribution of earnings are published. However, we utilise the unit data underlying the published survey results (see Appendix A).

The MCED data have a number of desirable features. First, the reliance on payroll information about employees' earnings avoids many of the reporting problems that arise when employees are surveyed directly.^[15] Second, tracing the pay for specific jobs rather than individuals provides a significant step towards pricing labour to constant quality.^[16] Third, the distinction between base pay and total pay permits an examination of wage flexibility for different classes of earnings. Fourth, and most useful, is the focus on the pay of those who have remained employed in a given position between surveys.

The measured earnings of stayers are unaffected by compositional change in the sample of respondents. For example, even with a fixed basket of jobs, compositional change often stems from changes in the experience of those who occupy the jobs. An increase in the share of experienced workers, who tend to have higher levels of earnings than less experienced workers, raises average earnings for the full sample, but has no effect on the average earnings of the stayers. Focusing on stayers also avoids the sampling problems that arise when employees who exit the survey have different characteristics and levels of earnings to those who replace them. Consequently, the reported changes in earnings should reflect actual wages growth rather than changes in the features of the sample.^[17]

Finally, the exclusion of positions characterised by award-only rates of pay removes a class of earnings that has an inherent institutional rigidity and comprises a diminishing share of wage income. Consequently, our analysis is confined to market-determined wages.

The MCED data are not, however, calculated on an hourly basis, so that if work effort varies substantially over time, some distortions in wages growth may arise.^[18] Furthermore, they are not based on a random sample but on a sample of firms that choose to participate in the survey, so that they can subscribe to information about prevailing wage rates. As it turns out, wage rates for skilled jobs in large firms are over-represented in the survey. But the coverage of the survey is, by any measure, broad and the number of observations is large, with a pooled sample of around 80,000 useable observations of annual wage changes. Consequently, useful inferences can be made about the distribution of wage changes in the market sector.

Footnotes

Primarily because validation tests showed that the individuals interviewed in the PSID reported different earnings than did their employers. [9]

In the same tradition as the PSID studies, wage rigidity has also been analysed using the British Household Panel Study, resulting in the recent and surprising claim that wage rigidity in the United Kingdom is far less pronounced than in the United States (Smith 2000). [10]

Until recently, this was also the case in Australia. For a general discussion of the various measures of average earnings see Reserve Bank of Australia (1996). [11]

For details of the design of the index see ABS (1998). In brief, pricing to constant quality requires removing from the index salary increments that are due to age, experience, work performance, change in qualifications etc, as these types of payments proxy measurable quality change. A change in the salary range for a selected job is treated as a ‘genuine’ price change and is recorded in the index. [12]

For example, the US employment cost index includes various payments that are clearly related to performance, and so does not price to constant quality. The New Zealand labour cost index does, however, price to constant quality and quantity in a manner similar to Australia's WCI. [13]

The Melbourne Institute has also developed a series of growth in wage rates that is loosely modelled on the WCI, but draws on less reliable household information. Furthermore, while it permits valuable insights into the characteristics of those wage earners at each point in the distribution of wage changes, it too is available over a relatively short period. See Melbourne Institute (1997, pp 4–5). [14]

Not only do validation studies reveal that employees tend to report lower levels of earnings than do their employers, they are less informed about the value of their non-wage remuneration. Furthermore, employees can claim to have remained in the same job when they have, in fact, remained with the same employer, or within the same industry, and performed different tasks. [15]

Although, failure to remove certain salary increments means that changes in market rates for specific jobs will, to some extent, reflect changes in the quality of labour. For example, in the MCED survey, base pay excludes increments that relate to performance, but not tenure. Yet changes in tenure (a proxy for experience) represent a change in the quality of labour. [16]

With the exception of wage cost indices, wage data are usually designed to provide an efficient estimator of the level of earnings rather than their growth. Since analysis of the distribution of wage changes requires an efficient estimator of wages growth, it is helpful to identify the stayers since they form a matched sample. A matched sample eliminates cohort variability – that is, the variation in measured wages growth that stems from comparing cohorts with different characteristics. We find that the reduction in cohort variability more than offsets the increase in sampling variability that arises from not using the full sample of respondents. For a review of these issues see Fuller (1990). [17]

In particular, if there is a rise in hours of unpaid work, our data will overstate growth in true wage rates and may mask some nominal wage falls. [18]