This week the Waikato Economics Discussion Group discussed this article by William Cochrane, Jacques Poot, and Matthew Roskruge, published in the Australasian Journal of Regional Studies (open access). This paper won the John Dickinson Memorial Award for the best paper published in AJRS last year. In the paper, Cochrane et al. look at the take-up of social security benefits in New Zealand territorial authorities (the lowest administrative level of government in New Zealand) as a result of two shocks: (1) the Global Financial Crisis (GFC) in 2008-09, and the Covid-19 pandemic in 2019-20. Interestingly, comparing those two shocks they note that:
...the initial impact of the GFC on social security benefit uptake was of a similar magnitude to that of the COVID-19 pandemic: a mean increase across TAs of 1.86 per cent versus 2.23 per cent respectively.
A small gripe is that those are actually percentage point increases, not percent increases. In other words, social security benefit uptake was 1.86 percentage points higher after then GFC than the year before, and was 2.23 percentage points higher during the Covid-19 pandemic than before. Importantly, not only is the increase in social security uptake similar for the two shocks, but the spatial distribution of the uptake of social security benefits is similar for the two shocks, as shown in Figure 2 in the paper:
The areas that are shown in darker blue had larger increases in social security uptake as a result of the shock (the GFC is the map on the left, and Covid-19 is the map on the right). It is clear from the figure that the shocks were more keenly felt in the North Island. In particular, Northland is heavily affected by both shocks, as well as the eastern Bay of Plenty and East Cape.
Cochrane et al. then turn their attention to looking at the factors associated with the increase in social security uptake, asking the question, what factors are associated with greater resilience (that is, what factors are associated with a lower increase in social security uptake). To do this, they rely on Census variables taken from the 2006 Census (three years before the GFC), and the 2018 Census (two years before the Covid-19 pandemic).
Since Cochrane et al. only have 66 observations of change for each period, and over 140 Census variables, this poses a bit of a problem. Cochrane et al. solve this issue in a few ways. First, they categorise their variables into 15 categories, and use just one variable in each category in separate cross-sectional regression models for each shock, and in spatial panel regression models that combine the data across both shock periods. Then, in a separate analysis they use a machine learning algorithm to select the most important variables for inclusion in the model.
The variables that are statistically significantly associated with social security benefit uptake vary somewhat between the models, but there are two variables that are consistently significant. First, territorial authorities that had a lower unemployment rate two years prior to the shock had lower benefit uptake. Second, territorial authorities that had a higher proportion of public sector employment had a lower benefit uptake. From the post-estimation regression model after machine learning, a one percentage point higher unemployment rate in the previous Census was associated with a 0.268 percentage point higher social security benefit uptake. A one percentage point higher public sector employment rate was associated with a 0.076 percentage point lower social security benefit uptake.
The implications of this (if we can interpret these effects as causal), is that if central (or local) government wants regions to be resilient to shocks, then finding ways of reducing unemployment (difficult) or increasing public sector employment (perhaps less difficult) are important things to consider. [*] However, as Cochrane et al. note in their conclusion, the current New Zealand government may actually be doing harm to resilience, because:
...if austerity measures were to be introduced in future years that lead to less public sector employment across all regions, either to reduce public debt or to fund tax cuts, our results do point to a likely decline in regional resilience.
*****
[*] An important consideration here is the definition of public sector employment. This isn't clarified in the paper (and I guess I could ask the authors, given that I know them all quite well, so I will). Table 1 in the paper tells us that public sector employment is, on average, about 14 percent. That is clearly more than just those included in the 'Public Administration and Safety' industry in the ANZSIC classification, which was about 5.4 percent of employment in the 2018 Census. But it is similar to the total of that category plus 'Health Care and Social Assistance', which was 14.9 percent of employment in the 2018 Census. However, if you were to include the health sector in public sector employment, why would you not also include 'Education and Training' as well (bringing the proportion to 23.0 percent in the 2018 Census)?
No comments:
Post a Comment