An illustration that using data for a conclusion is not objective

There is an interesting post over at John Taylor’s blog (ht Greg Mankiw). In it he shows that government spending as a % of GDP has a strong positive correlation with unemployment, and that investment as a % of GDP has a strong negative correlation.  His conclusion is to focus on cutting taxes to increase investment instead of spending.

Update:  Here is the point I’m trying to make written up more clearly and completely.

Now, this way of viewing the data is consistent with the model he has in his head – but it is only by stating that model that we can look at the value judgments involved to figure out what we agree with or disagree with.

Looking at the data alone, we cannot make this conclusion – we can just say there are correlations.

For example, I would note that investment responds disproportionately to the economic cycle – this is a well known “stylized fact”.  The excuse economists often use is that businesses and households cut back on durable good expenditures most heavily when we enter a downturn – as the durable products they already own act as effective substitutes for new durables.

As a result, assume that government spending is a constant (so the government is doing nothing to smooth the economic cycle).  When GDP falls, investment falls more steeply.  When GDP falls, unemployment rises.  In this case, even with no government smoothing, the existence of an economic cycle will lead to BOTH of the observed correlations above (note that GDP is a denominator) – there is no way we can reach any policy conclusion from them.

We need a model, with a counterfactual – then we can use the data, and value judgments, to reach policy conclusions.  His model says that these correlations are causal.  My model would probably say that all these correlations suffer from too much endogeniety, and I would state that appropriate monetary policy is the best way to move forward – as it would reduce the observed variability in investment, unemployment, and GDP.  Both conclusions use the same data, the underlying models are just a bit different.

He is of course world famous and ridiculously intelligent, while I’m an arbitrary blogger – but I still prefer my conclusion, that’s what value judgments are for right 😉

2 replies
  1. Matt Nolan
    Matt Nolan says:


    Very good. I was surprised Krugman didn’t try to attack it more.

    Tbf to Taylor from the Freakonomics criticism – it is common to look at post-1990 data, because a lot of people believe there was a “structural break”. The new phillips curve people do the same thing – which is why Krugman didn’t state it.

    The main thing for me is definitely interpreting the data – I’m used to looking at those figures and the policy conclusion of “cutting corporate tax rates” wasn’t actually the first thing that popped into my head. The first thing that popped into my head was “and what, this is what mainstream theory says would happen – its tells us nothing about the impact of policy”.

Comments are closed.