Beyond the averages: Measuring underlying wage growth using Labour Force Survey microdata

Introduction

Wage growth is a key indicator that central banks monitor because labour costs are an important component of production costs and inflation. However, average wage growth can be a misleading measure of inflationary pressures. This is because it is a simple average of the wages earned by millions of people who have different skills, levels of experience and occupations. As a result, the average wage fluctuates based not only on labour market conditions but also on the composition of the workforce.1 However, wage growth driven by compositional changes is generally not informative about inflationary pressures and creates fluctuations that should be looked through.

At no time was this more apparent than during the COVID-19 pandemic (Chart 1). In April 2020, when lockdowns were in place in much of the country, the average hourly wage was $31.14, up about 7% from two months earlier. Yet individual workers saw barely a change in their earnings during this short period. Instead, the jump in the average wage was primarily because job losses during the lockdowns disproportionately affected low-wage workers in high-contact industries, such as accommodation and food services. With fewer low-wage workers in the workforce temporarily, the average wage rose dramatically.

The challenge of measuring wage growth

Economists and statisticians have used several techniques to obtain a measure of wage growth that is adjusted for changes in the composition of the workforce. The most accurate and straightforward approach involves using specialized surveys that directly control for compositional changes. A prominent example is the Employment Cost Index (ECI) from the US Bureau of Labor Statistics. The ECI measures the cost of employees to employers by tracking changes in wages and benefits and controlling for the effects of workers switching jobs or industries. Another technique involves using microdata to track the wage growth of individuals. For instance, Daly, Hobjin and Wiles (2012); Almuzara, Audoly and Melcangi (2023); and the Federal Reserve Bank of Atlanta’s Wage Growth Tracker use results from the US Bureau of Labor Statistics’ Current Population Survey to extract the median year-over-year wage growth of individuals who are observed both in the current month and 12 months prior. In a more involved approach, Bils, Kudlyak and Lins (2023) use work histories from the long panels in the US National Longitudinal Surveys of Youth to keep constant the quality of the matches between workers and firms.

Unfortunately, these techniques are not possible in Canada, partly because there are no specialized surveys similar to the ECI. Additionally, the design of Statistics Canada’s Labour Force Survey (LFS) does not allow for tracking the wages of specific individuals over a sufficiently long period of time. Respondents to the LFS are interviewed over the course of only six months. And they are asked about their wages only either in the month when they enter the survey or if they change jobs.

A useful wage growth measure should:

• give a clear signal without much volatility
• reflect the balance of supply and demand in the labour market
• be informative about inflationary pressures
• be timely

A new wage growth measure

To achieve these goals, we introduce a new measure of underlying wage growth that we call LFS-Micro. This measure works within the limitations of the publicly available microdata from the LFS to calculate a measure of underlying wage growth that is separate from shifts in the composition of the Canadian workforce. We believe this approach improves our ability to monitor inflationary pressures coming from the labour market.

We find that composition-adjusted wage growth averaged 3.9% between January and August 2024. While this is still elevated relative to its history, it is significantly below the average wage growth of 5.1% recorded over the same period. Additionally, we find that a change in the occupational makeup of the workforce is the largest compositional effect, mostly due to an increase in the prevalence of high-earning management jobs.

Data and methodology

The LFS is the timeliest source of data on Canadian wages. The survey uses a rotating panel design where each responding household is surveyed monthly for six consecutive months. In the first month, all employed respondents are asked about their wages. However, in subsequent months, respondents are asked about their wages only if they report specific changes to their employment information.2 Otherwise, respondents are assumed to have the same wage as in the previous month. This might create a lag in the wage data. In this analysis, we define wages as the usual hourly earnings of employees at their main job before taxes and deductions and including tips and commissions.

Statistics Canada maintains two versions of the LFS microdata:

• a fully anonymized version available for public use, known as the public use microdata file (PUMF)
• a full analytical version, known as the master files, only available for use within Statistics Canada’s secure environment

Both types of files include the main survey variables and the weights required to reproduce key statistics, including average wage growth. Importantly, the PUMF does not allow for respondents to be identified longitudinally and is missing some details such as the exact age of respondents and details about their immigration status.3

We use the PUMF in this analysis because it is timely and publicly available, which allows other users to reproduce our results. We also provide in this note the results of a robustness check with estimates using the confidential LFS data.4

Our goal is to estimate underlying wage growth, which is total wage growth when keeping constant the composition of the workforce. While panel data techniques would be natural candidates for achieving this goal, they are not applicable to the PUMF. Instead, we use one of the main tools for disentangling wage growth from compositional changes in cross-sectional data: the Oaxaca-Blinder decomposition. This methodology has already been used in similar analyses (Christodoulopoulou and Kouvavas 2022; Aizcorbe and De Haan 2024).

The Oaxaca-Blinder decomposition is frequently used in labour economics to identify differences in wages that can be attributed to demographic differences across groups of workers. For our purposes, we look across months instead of groups, estimating period-by-period regressions of wages on workers’ characteristics. This technique allows us to separate the portion of wage growth that is due to changes in the prevalence of those characteristics in the workforce from increases in quality-adjusted labour costs, which represent increases over time in the cost of equivalent units of labour.

The latter is what constitutes our estimate of underlying wage growth isolated from compositional changes: LFS-Micro.

Wage growth can be written as follows, where \(W\) is the log of wages:

\(\displaystyle∆E(W_t )\) \(\displaystyle=\,E(W_{t+1} )\) \(\displaystyle-\,E(W_t )\)

The decomposition method consists of regressing the log of wages \((W)\) on a set of observable characteristics \((X)\):5

\(\displaystyle W_t\) \(\displaystyle=\,X_t^{'} B_t\) \(\displaystyle+\,e_t\)

Plugging this into the wage growth expression:6

\(\displaystyle ∆E(W_t )\) \(\displaystyle=\, \frac{[E(X_{t+1})-E(X_t )]^{'} B_t}{Composition\ Effect} \) \(\displaystyle+\, \frac{E(X_t )(B_{t+1}-B_t )}{Underlying\ Wage\ Growth} \) \(\displaystyle+\, \frac{[E(X_{t+1}^{'}-E(X_t^{'} )](B_{t+1}-B_t)}{Interaction\ Term}\)

The identity shows that wage growth can be decomposed into:

• the sum of compositional changes
• changes in the returns to characteristics, or underlying wage growth—our Micro-LFS measure
• an interaction term

The interaction term is very small and is excluded from our results.

This decomposition not only produces a measure of wage growth adjusted for compositional changes but also provides insights into shifts in workforce composition and how they affect the observed wage growth.

Results

Compared with other measures, LFS-Micro provides a much smoother estimate of wage growth that is more representative of underlying wage pressures (Chart 2). In general, LFS-Micro wage growth is lower than the LFS average wage growth measure over the full sample period. This suggests that compositional changes in the labour market over the past few decades have generally contributed positively to wage growth.7

Two periods are particularly important to our understanding of recent wage dynamics:8

Wage growth in 2020: In the early phases of the COVID-19 pandemic, average wage growth spiked, peaking at over 10% in April 2020. This was likely because the majority of layoffs at the time affected low-wage earners, raising the average wage. Since LFS-Micro controls for changes in occupation and industry make up, it remains relatively flat during this period, likely giving a more accurate measure of wage growth at the time for workers who remained employed.

Wage growth since 2022: In the summer of 2022, average wage growth began to climb well above its historical range due to a combination of factors, including an extremely tight labour market and high inflation expectations. LFS-Micro is notably softer than average wage growth during this period. This suggests that while wage growth was quite elevated, the average measure has generally exaggerated the strength of growth in labour costs.

The most important compositional changes

A benefit of LFS-Micro is that compositional changes can be analyzed to assess their importance for average wage growth. Chart 3 shows the decomposition of wage growth due to compositional changes since January 2019. Most upward pressure in average wage growth both in the first year of the pandemic and since mid-2022 comes from changes in the occupational makeup of the labour market. Specifically, this pressure relates to an increasing share of people working in high-wage occupations.

This shift to high-wage occupations in April 2020 was because low-wage workers in front-line sectors were disproportionately affected by layoffs at the onset of the pandemic. Starting in mid-2023, the tilt toward high-paying occupations stemmed from an increase in jobs classified as management. Changes in the average educational attainment of workers also played a notable role during both periods because of selective layoffs. Finally, industry and tenure seem to matter during the pandemic, with the former likely responding to structural adjustments due to public health restrictions and the latter to layoffs of recently hired employees. Note that these results do not total the difference between LFS-Micro and the average wage growth due to the log-transformation of wages in the regression.9

The new wage measure compared with other measures

The Bank of Canada monitors a number of measures to inform its understanding of wage growth. Table 1 shows how LFS-Micro compares with seven other wage measures.10 Based on several important benchmarks, LFS-Micro either outperforms or performs well compared with the current suite of measures. This makes our measure a useful tool to gauge inflationary pressures stemming from the labour market.

• Volatility: The standard deviation of each wage measure is typically inversely related to the signal-to-noise ratio, which is the amount of meaningful information relative to the amount of non-relevant data. Quarterly wage growth for LFS-Micro is the least volatile among the eight measures, suggesting it provides a more reliable signal.
• Relationship with labour market conditions: The unemployment rate is one of the main indicators of excess supply in the labour market, while the vacancies-to-unemployment ratio can indicate excess demand. Among the eight wage growth measures, LFS-Micro correlates most strongly with both the unemployment rate and the vacancies-to-unemployment ratio. This suggests that LFS-Micro is a good measure of wage movements that are bringing the labour market into balance. Additionally, we tested our wage measure’s link to multiple wage growth drivers simultaneously by including it in the Bank’s Canadian replication of Bernanke and Blanchard’s (2023) model of inflation, prices and wages (Bounajm, Junior Roc and Zhang 2024). The results show that the measure has a stronger fit (R-squared) with respect to relevant macro variables.
• Relationship with inflation: Theoretically, changes in wage growth should be closely related to changes in inflation because labour is a key production input. LFS-Micro has a substantially strong correlation with measures of inflation. In particular, it shows the strongest correlation with services excluding shelter, which is a component of the consumer price index (CPI) basket of goods and services that is usually labour intensive.
• Timeliness: The LFS is the most timely official survey of the labour market.

Disaggregate measures of LFS-Micro

In general, the LFS-Micro technique can be applied to any segment of the labour market to obtain a composition-adjusted estimate of wage growth. Using microdata allows us to filter for workers with specific demographics, including public and private sector workers. The results for these groups, displayed in Chart 4, show that:

• the private sector has largely driven average wage growth since 2022
• public sector wage growth has been rising more gradually since 2023

Notably, despite the elevated wage growth in the private sector, the level of wages remains higher in the public sector.

Because wage growth can vary significantly across different groups of workers, the flexibility of LFS-Micro improves the Bank’s ability to understand the breadth of underlying wage growth. Other examples include looking at wage growth by industry or gender. However, applying this technique to small segments of the labour market may be difficult because of an insufficient number of observations.

Robustness checks using microdata from the master files

The main results in this note have been produced using the PUMF because of its timeliness and to allow other users to reproduce our results. However, it is worth considering whether the extra detail in the LFS master files could improve our estimates, particularly with respect to the correct identification of respondents’ immigration status.11

Specifically, non-permanent residents (NPRs) are not identified as migrants in the PUMF. Instead, NPRs end up in the same category as Canadian-born workers. Given the increase in the number of NPRs in Canada since mid-2022, we consider the extent to which the lack of identification of NPRs in the PUMF causes a bias in the wage measure.12

To answer this question, we calculated LFS-Micro wage growth using both the PUMF and the master files (Chart 5). The results show that using the PUMF does not cause major discrepancies relative to using the master files. However, wage growth calculated using the PUMF in the last two years is slightly below the wage growth from the master files. This is a direct consequence of the PUMF not classifying NPRs as migrants. NPRs generally receive lower wages than the broader population, causing a downward bias in the LFS-Micro wage measure. However, the differences are very small even in the last two years, which is consistent with the small size of the NPR workforce.

Chart 5: Taking advantage of detail in Labour Force Survey master files does not materially affect results

Chart 6 breaks down the factors causing wage growth to deviate from the underlying wage growth. Unlike in Chart 3, we include a set of immigration status indicators that separate NPRs from Canadian-born workers. We confirm that immigration status has a negative effect on average wage growth that is larger than what the data in the PUMF suggest.

Overall, taking advantage of the extra detail in the master files does not substantially affect the results. This confirms that using the PUMF is adequate for calculating underlying wage growth—which is convenient given its ease of access and timeliness.

Appendix

Table A-1: Evaluating various wage growth measures across multiple criteria

Note: LFS is Labour Force Survey. SEPH is Survey of Employment, Payrolls and Hours. ULC is unit labour costs. See Bernanke and Blanchard (2023) for more about the authors’ wage equation. Sample period is 1998Q2 to 2019Q4. More information on these measures can be found on the Bank's website.

References

Aizcorbe, A. and J. de Haan. 2024. “An Application of the Oaxaca-Blinder Decomposition to the Price Deflation Problem.” US Bureau of Economic Analysis Working Paper No. 2024-2.

Almuzara, M., R. Audoly and D. Melcangi. 2023. “A Measure of Core Wage Inflation.” Federal Reserve Board of New York Staff Report No. 1067.

Bernanke B. S. and O. J. Blanchard. 2023. “What Caused the US Pandemic-Era Inflation?” Paper prepared for the Hutchins Center on Fiscal & Monetary Policy at the Brookings Institution conference, “The Fed: Lessons Learned from the Past Three Years,” May 23.

Bils, M., M. Kudlyak and P. Lins. 2023. “The Quality-Adjusted Cyclical Price of Labor.” Journal of Labor Economics 41 (S1): S13–S59.

Bounajm, F., J. G. Junior Roc and Y. Zhang. 2024. “Sources of Pandemic-Era Inflation in Canada: An Application of the Bernanke and Blanchard Model.” Bank of Canada Staff Analytical Note No. 2024-13.

Brochu, P. 2021. “A Researcher’s Guide to the Labour Force Survey: Its Evolution and the Choice of Public Use Versus Master Files.” Canadian Public Policy / Analyse de Politiques 47 (3): 335–357.

Christodoulopoulou, S. and O. Kouvavas. 2022. “Wages, Compositional Effects and the Business Cycle.” European Central Bank Working Paper No. 2653.

Daly, M. C., B. Hobijn and T. S. Wiles. 2012. “Dissecting Aggregate Real Wage Fluctuations: Individual Wage Growth and the Composition Effect.” Federal Reserve Bank of San Francisco Working Paper No. 2011-23.

Kirubaharan, E. and A. Bernard. 2024. “Earnings and Wages – A Guide to Using Indicators from the Survey of Employment, Payrolls and Hours and the Labour Force Survey.” Statistics Canada Labour Statistics: Technical Papers, Catalogue No. 75-005-M.

Endnotes