Submission 167 Australian Council of Trade Unions Workplace Relations Framework Public inquiry

A research vacuum

In the broader debate about whether minimum wages are an ‘anti-poverty measure’,¹⁸⁶ Neumark suggests that the question of whether minimum wages ‘destroy jobs’ is crucial: if they don’t then they are a ‘free lunch’ that helps reduce poverty.¹⁸⁷ While Neumark argues that evidence from many countries indicates that minimum wages reduce the jobs available to low-skilled workers, this claim is strongly contested. This field of research remains a contested one and it is not clear how one could ever arrive at definitive results because there are no universal economic truths in the labour market, only historically contingent outcomes which reflect the institutional arrangements prevailing in local labour markets.

Much of this international debate in the English-speaking countries has focussed on either the US labour market or the situation in the UK. In both cases, their historical, geographical and institutional arrangements differ markedly from that which prevails in Australian. In the US different states have different minimum wages, and they are set at different times. As Neumark showed, between 2004 and 2008 the number of US states which had minimum wages higher than the federal minimum wage grew from about 12 to over 30. In other words, there were regular adjustments in wage levels by different jurisdictions. Such variation in state minimum wages lends itself to research approaches such as the ‘natural experiments’ method, where one contrasts employment outcomes between one jurisdiction which obtains a pay increase and one which doesn’t. In the case of the UK, one can also contrast a period without any minimum wage to a period when the minimum wage applied.

In both cases, large numbers of workers are on the minimum wage and one might therefore hope to see any potential effects surfacing in the wider labour market. In Australia, there have been attempts to use this ‘natural experiments’ approach, but they have foundered on the problem that most minimum wage workers are covered by the Federal system and that state variations are minor. In 2009, when there was no increase in the FMW, New South Wales broke ranks and awarded an increase. But the number of workers affected was too small to be provide a useful test case. One study, carried out by Andrew Leigh in 2003, also came to grief on this problem of small numbers.¹⁸⁸ Leigh used a difference-in-difference estimator to analyse changes in Western Australia between 1994 and 2001, but his large estimates—initially calculated on the wrong data, but subsequently reproduced using the correct data—led to scepticism by many. As Neumark and Wascher observed:

The elasticities that Leigh reports for aggregate employment of young individuals are quite large relative to those found for other industrialized countries, especially given his estimate that only about 4 percent of workers were affected by these changes in the minimum wage. Unfortunately, he does not offer a potential explanation for the size of his estimates, and in the absence of such an explanation, the magnitudes of these estimates, at least, might be regarded sceptically.¹⁸⁹

In other words, with only four percent of workers affected by these changes, the likelihood of discerning genuinely large elasticities amidst the noise seems remote. It is important to keep in mind the Bayesian critique of frequentist approaches: that in any modelling of random data, one will regularly find statistically significant results, even when there is nothing there but noise.¹⁹⁰

More seriously for Leigh’s endeavour was the problem of the ‘control group’. Where the US studies compared different states, and ideally comparable states, and the UK studies compared different time periods for the same labour market, Leigh’s Western Australian study had no similar comparator. Instead, he used ‘the rest of Australia’ and simply assumed that his difference-in-difference estimator would remove all confounding. In other words, because he was comparing differences with differences he assumed he did not need to take account of other ways in which Western Australia differed from the rest of Australia. As Machin and Manning have argued, the use of macro models like that used by Leigh assume that all sectors of the labour market behave in the same way. Not only is this unlikely within a single state, but it’s highly unlikely across the country. ¹⁹¹

• 
  The conditions necessary to undertake a ‘natural experiments’ study in Australia to explore minimum wages and employment would need to satisfy a number of conditions:
  
• 
  the number of workers affected would need to be substantial;
  
• 
  one would need to take account of the pay scales, the way the FMW filters through a range of award rates at different levels;
  
• 
  the jurisdictions to be compared would need to be genuinely comparable, and not attempt to use a ‘rest of Australia’ artefact;
  
• 
  one would need repeated instances of the changes over time, so that the effects could be discerned from the noise and one could be confident that one had isolated a genuine relationship;
  
• 
  one would need to model time lags adequately, and take account of the complexity of award flow-ons at the state level (where this still happens).
  

Posed in this way, the task is daunting. The kinds of data necessary for this task are formidable. ABS data, in the form of the EEH, allows one to examine the industrial instruments which cover workers, but this does not take account of the pay scales issue and the different levels at which the FMW ‘cuts in’. As an employer-based survey, EEH has many advantages, but tracking individuals over time is not one of them. Therefore, one would need to look for macro-economy effects from any exercise using these data (an approach regarded with some skepticism by Healy and Richardson).¹⁹²

If one chose instead to use household unit record data, such as HILDA, to try to track a cohort of minimum wage workers likely to have been affected by FMW increases, then other problems arise. For several waves, the HILDA team implemented a question on industrial instruments but they soon abandoned this as unreliable. Consequently, while attempts have been made to use HILDA to track minimum wage workers, this has primarily been done by dividing employees into those sitting ‘At or below the FMW’ and others, using an estimate of the hourly rate.¹⁹³ Some more fine grained approaches have been used with the HILDA data, again based on hourly rate assumptions, such as distinguishing between those on the C10 rate and an upward limit ($700 per week in 2007), those above the FMW and below the C10 rate, and those on or below the FMW.¹⁹⁴ Studies such as these have been useful for descriptive purposes, and for modelling broad labour market transitions, but their utility for modelling elasticities of labour demand is problematic.

One recent study by Alex Olssen used the HILDA data and avoided some of these difficulties by concentrating on youth rates, where the pay scales are closely tied to age. That is, as juniors reach a particular age, their hourly rate—as specified in the award—increases to a set amount. Using a regression discontinuity design he exploits this variability in earnings tied to age, by assessing whether employment levels differ according to such variability. One of the advantages of Olssen’s approach was that the coverage was extensive: between 70 and 90 per cent of youths in the food industry earned wages at or below the award minimums. While Olssen’s approach was an innovative one, it did rely on purely local and short-term effects.¹⁹⁵