One of the difficulties in economics is isolating the effects of particular actions in a very complex world. If we cut income tax this year and next year tax revenues are a little higher, it’s tempting to attribute that to the tax cut. But maybe it’s actually because oil prices fell, or because growth was picking up anyway.
To get around this, economists try to aggregate large numbers of data points – that is, look at lots of different times when we cut income tax and see what the effect on revenues were, ideally adjusting for things we know might affect growth, like the price of oil. Using lots of different data points helps us to cancel out ‘noise’ and focus in on the effects we really care about.
Another example: If oil prices rise, and consumption of oil doesn’t fall, we don’t throw out our model that says people use less oil when prices rise—we acknowledge that oil’s price isn’t the only factor. Perhaps it was particularly cold just after oil prices rose, so people needed more for heating; perhaps there was a big national holiday and everyone used their car more. Similarly when the price of labour jumps and employment doesn’t fall, this doesn’t mean that employers don’t take wages into account, it might mean there are countervailing factors: employers can pass some costs on; employers can reduce other benefits; or employers are going to reduce hiring to take account.
This is why it can be foolish to point to a single example of a tax cut appearing to raise revenues or to point to a single example of raising or introducing a minimum wage not causing unemployment. The latter has been very common recently. We didn’t see unemployment rise in 1998 when we introduced the National Minimum Wage, so people saying the new National Living Wage will hit jobs are on thin ice.
The LSE’s Prof Alan Manning, who is an expert on minimum wages, does this in the FT today, being quoted as saying that “prophecies of doom … turned out to be wildly inaccurate then; I suspect they will be this time as well.” I guess Prof Manning is being glib – he knows all the literature and that a single event isn’t indicative of very much, but it’s misleading to most people reading who do not.
The graph above, via Menzie Chinn, shows a meta-analysis of the impact of minimum wage rises on employment from 1,424 data points – an elasticity of -0.5 means that a 1% rise in the minimum wage is associated with a 0.5% fall in employment for the affected group. The red line is the mid-point of a range (-0.1 to -0.3) suggested by David Neumark. The graph above seems to suggest that the effect is negative but small; Neumark argues that the international evidence points to a clear disemployment effect.
Other research, which is again based on data from large numbers of events, not a single event, suggests that minimum wage rises tend to slow down job creation rates over time. The effect here seems to be that employers are reluctant to actually fire workers (perhaps for the same reasons they are reluctant to do so during recessions) but become less willing or able to hire new ones. Another paper suggests that job losses are avoided by passing the costs on to consumers.
The right is guilty of the kind of error I'm criticising, too – the OBR disputes the claim that cutting the 50p tax rate really raised revenues, suggesting that the increased revenue came from people deferring income until the rate was cut. I don’t hear many supporters of the 50p cut acknowledging that.
It also needs to be pointed out that the level, not just the rate, of the rise in the minimum wage matters too. In 1998 the NMW was introduced at £3.60 per hour, or £5.71 in today’s prices; the new National Living Wage will be £7.20 per hour. A comparably small rise may still raise the level to a high enough point that it does cause serious problems in terms of job losses.
It may be that the NLW does cause job losses, which are masked by other positive effects. It may be that it doesn’t, but the economy dips anyway and it looks as if it does. It will be impossible to say either way if we just look at this one event. The trick is to look at many events and test our hypotheses against the aggregate, not to cherry pick single events to make a point.