Meta-Analysis of the Impact of Fiscal Policies on Long-Run Growth

March 2003
261 citations

Peter Nijkamp
Jacques Poot

original article

This is a boring summary of a study. The "Detailed Summary" section aims at summarizing precisely what the study's authors said with no additions on my part. The main benefit is that this post is an order of magnitude more concise than the original.

High-Level Summary

  1. High-Level Summary
  2. Detailed Summary
    1. Introduction
    2. Data
    3. Exploratory Analysis
    4. Government Size
    5. Defense Spending
    6. Taxation
    7. Public Infrastructure
    8. Education
    9. National v. Regional
    10. Conclusion
  3. Commentary
Effect of Taxation and Spending On Economic Growth
Positive Impact Negative Impact Inconclusive
Education Spending 11 1 0
Infrastructure Spending 28 3 8
Taxation 0 6 4
Defense Spending 1 11 9
Overall Spending 7 12 22

There is strong empirical support that education and Infrastructure spending boost long-run economic growth. There is moderate support for defense spending and taxation having negative effects on long-run growth. The effect of government spending as a proportion of GDP on growth seems minimal. Sadly, this study didn’t study effect-size.

Detailed Summary

Introduction

In this paper, Nijkamp and Poot study the relative impacts of different fiscal policies on long-run economic growth. More precisely, they divide fiscal policy into five areas:

  1. General government consumption
  2. Tax rates
  3. Education spending
  4. Defense spending
  5. Public infrastructure spending

They find that education and infrastructure spending are important to long-run growth, but that other forms of government spending are less obviously constructive.

Data

In 1986, Landau noted an “extensive literature search turned up only three papers.” Nijkamp and Poot likewise found this to be the case, so all of their 93 journal articles were published between 1983 and 1998. Since the late 90s, new research has dwindled. Nijkamp and Poot tried to balance coverage (i.e. avoiding publication bias) and precision (i.e. using good studies).

To ensure precision, they only used studies

  1. from Barro’s “Determinants of Economic Growth” (1996) (5181 citations)
  2. from Dunne’s Peace Dividend (1996)
  3. that were refereed journal articles

“Pure theory” articles, non-English articles, and “hard to retrieve” articles were excluded. Nijkamp and Poot claim that they include most of the relevant articles between 1983 and 1998 with the exception of some studies on taxation and education. They also note that they found only 10 articles on the effects of taxation, but that more recent work is available for later analysis.

Because some of these 93 studies examined more than one of the five policy areas and some used multiple data sets, they gave Nijkamp and Poot 123 observations.

Exploratory Analysis

94% of the studies accounted for population growth. 81% controlled for the investment or savings rates (which almost always correlates positively with growth). 56% accounted for the fact that high income countries grow slower than low income ones – this choice affected the results (discussed later).

One limitation of this analysis is that it more heavily focuses on developed nations (where the most solid data is available for study), so it’s unclear how much we should generalized to developing ones.

Finally, 82% of the observations focused on national government spending; however, it analysis does not reveal a systematic difference between the effects of national- and regional-level spending.

Government Size

41 studies in the sample considered overall government size. 90% did not count financial transfer payments (e.g. social security, welfare, and subsidies) as part of government spending and used spending as a ratio of GDP as their measurement of government size. The remaining four did include transfers.

Of these 41, 29% concluded that larger governments reduce growth, 17% concluded the opposite, and the remaining 54% were inconclusive. Thus, we conclude that aggregate government spending has “no clear impact on long-run growth at the macro level.” Moreover, the conclusion each study came to appears to be independent of size, year of data, year published, or whether national or regional data was used. Nijkamp and Poot did, however, find that studies were less likely to find a negative effect of government spending on growth in developed economies, but were more likely to find an effect in cross-section studies – a feature, perhaps, of

  1. Wagner’s Law, which states that as countries get wealthier a greater and greater proportion of their economy is devoted to government spending
  2. The fact that developed nations tend to grow slower than developing ones

Defense Spending

Of the 21 studies that examined defense spending, 1 suggested that defense spending boosted growth, while 11 concluded it was detrimental. Finally, rough set analysis also concluded that the studies finding a positive impact of defense spending on growth tended to be older, smaller, and focused on developing economies.

Taxation

Phillips and Gross completed a meta-analysis in 1995 on the effect of state and city taxes in the United States. They found that taxes had a limited negative effect on state growth, but a much larger effect on city growth. Curiously, they did not find different tax elasticities.

In Nijkamp and Poot’s sample, there were 10 studies that examined the effect of taxes on growth. None found that taxes boosted growth, but 60% found it reduced it. However, the small sample of 10 studies make it difficult to draw definite conclusions.

Public Infrastructure

Of the 39 observations that studied public infrastructure, 72% found a positive impact, while 20% were inconclusive.

A meta-analysis by Button in 1998 found that when smaller regions engaged in public infrastructure investment, they received smaller economic benefits, probably because nearby regions reaped some of the reward. This conclusion was supported by Nijkamp and Poot’s analysis.

The mean time span is our sample is 28 years, and it appears (p=10%) that studies over longer timespans are more likely to find a significant relationship between investment in infrastructure and growth.

Rough set analysis found that studies that find infrastructure spending increases growth tend to

  1. be based on time-series.
  2. be published in unranked journals.
  3. use older data.
  4. have smaller datasets.

Education

11 of the 12 observations support the theory that education positively affects growth, and Nijkamp and Poot believe the one non-positive study was due to too strict a significance test. Rough set analysis found that most studies on education had smaller datasets than the other areas.

National v. Regional

Using rough set analysis, Nijkamp and Poot found that studies could be correctly classified without reference to whether the data was national or regional, supporting the idea that we can use conclusions from the former (from which data is plentiful) to the later (where data is less so).

Conclusion

While evidence supported the theory that education and infrastructure investment boosted long-run economic growth, evidence for the others was rather weak. However, the results depend a great deal on the type of data and analysis. Studies published at top-tier journals are less likely to reject the null-hypothesis of no-effect.

Commentary

First, I want to show you the table again:

Effect of Taxation and Spending On Economic Growth
Positive Impact Negative Impact Inconclusive
Education Spending 11 1 0
Infrastructure Spending 28 3 8
Taxation 0 6 4
Defense Spending 1 11 9
Overall Spending 7 12 22

I just want to talk a moment about the inconclusive studies. While inconclusive studies are evidence that an effect is not strong, they really shouldn't change our beliefs regarding whether the net effect is positive or negative.

So, if our goal is to determine whether a type of spending has a positive or negative effect, then it makes sense to look at the proportion of positive to negative studies and figure out whether we can reject the null-hypothesis.

If we actually carry out this analysis, we find that each of these p-values are below 2% (in the direction with more studies), with the exception of the Overall Spending category, which has a p-value of 9.6%.

Under this interpretation then, we have solid reason to believe that education spending and infrastructure spending improve long-run growth; taxation and defense spending decrease it; and that overall spending has an ambiguous (possibly negative) effect.

Of course, this all ignores the quality of the studies and the size of the effects found in them.