What drives health care expenditure?—Baumol's model of ‘unbalanced growth’ revisited

Abstract

The share of health care expenditure in GDP rises rapidly in virtually all OECD countries, causing increasing concern among politicians and the general public. Yet, economists have to date failed to reach an agreement on what the main determinants of this development are. This paper revisits Baumol's [Baumol, W.J., 1967. Macroeconomics of unbalanced growth: the anatomy of urban crisis. American Economic Review 57 (3), 415–426] model of ‘unbalanced growth’, showing that the latter offers a ready explanation for the observed inexorable rise in health care expenditure. The main implication of Baumol's model in this context is that health care expenditure is driven by wage increases in excess of productivity growth. This hypothesis is tested empirically using data from a panel of 19 OECD countries. Our tests yield robust evidence in favor of Baumol's theory.

Introduction

The share of current health care expenditure (HCE) in the gross domestic product (GDP) rises rapidly in virtually all developed nations. Fig. 1 illustrates this for a couple of countries for which data are available back to 1960. According to the OECD's health database 2005, only two countries (Denmark and Ireland) record similar shares as 20 years ago. In the U.S. on the other hand, the share has risen from 4.8% in 1960 to 14.7% in 2003. This is the highest value of all OECD nations. Switzerland ranks second in this statistic with a share 11.3% in 2003 (up from 4.8% in 1960). An opinion poll carried out in this country in early 2006 revealed rising ‘health care costs’ to be the most worrisome topic for Swiss households (cf. NZZ, 2006).

In view of this large public concern, it is unfortunate how little we know about the factors that drive the rapid rise in HCE. Back in 1994, Hoffmeyer and McCarthy (1994, p. 67) wrote that “there is just one, very clear, very well-established statistical fact relating to health care expenditure: its correlation with GDP. No other robust and stable correlations have yet been found.” This statement is confirmed by Roberts (1999) who dates the starting point of cross-country research into the determinants of HCE back to Newhouse (1977) and then writes: “During this time [the past 20 years] there has been little progress beyond the finding that variations in per capita national income are closely correlated with variations in per capita health spending” (Roberts, 1999, p. 459).

In the literature this judgment is based on we can distinguish two stages. Between the mid-1970s and the mid-1990s, scholars such as Kleiman (1974), Newhouse, 1977, Newhouse, 1987, Cullis and West (1979), Leu (1986), Parkin et al. (1987), Culyer (1990), Milne and Molana (1991), Getzen and Poullier (1991), Gerdtham and Jönsson, 1991a, Gerdtham and Jönsson, 1991b, Gerdtham et al. (1992), and Hitiris and Posnett (1992) provided evidence for a positive correlation between HCE and GDP (mostly) in OECD data. This correlation was found to be robust to varying years covered, estimators and use of conversion factors (such as deflators, exchange rates, or health care purchasing power parities). Other intuitively plausible explanatory variables were normally not found to be statistically significant.¹ An important issue in this first stage of research was the question whether health care is a ‘luxury good’, i.e. whether a larger than proportionate increase in income is spent on health care (cf. Getzen, 2000).

In the more recent stage of research – which started with Murthy and Ukpolo (1994) and Hansen and King (1996) – the time series properties of the variables in question have received more attention than before.² Unit root and cointegration tests were performed for HCE and GDP. The results of these tests have been somewhat inconclusive and not very robust to the choice of the testing methodology. Hansen and King (1996), for instance, found no cointegration between HCE and GDP for all OECD countries except Iceland applying the Engle–Granger (EG) two-step method and testing the residuals from the cointegrating regression with the Augmented Dickey Fuller (ADF) test. Blomqvist and Carter (1997) reversed the result using the Phillips–Perron (PP) test for the second step. There is a related controversy on whether the variables are non-stationary in the first place. McCoskey and Selden (1998) reject the null hypothesis of unit roots for HCE and GDP while Roberts (1999) finds both variables to be non-stationary. Based on country-by-country as well as panel cointegration tests, Gerdtham and Löthgren (2000) confirm that both health expenditure and GDP have a unit root and that they are cointegrated. Yet, as the latest twist in this debate come the contributions by Jewell et al. (2003) and Carrion-i-Silvestre (2005) who, for the first time in this body of literature, consider the possibility of structural breaks in the time series. Using data from 20 OECD countries, they find both HCE and GDP to be stationary around up to five structural breaks.

Reviewing the literature, we cannot help but admit that despite intense research efforts, we do not really know the degree of integration of health expenditure variables. Unfortunately, given that the available time series are rather short, which lowers the power of the tests, and that the number of competing tests is huge (and growing), some uncertainty is likely to remain with respect to the properties of the time series analyzed in this field of research.

A fresh approach might be attractive not only for methodological, but also for theoretical reasons. Over the past 30 years, research into the determinants of health care expenditure has concentrated on – and to some degree even confined itself to – evaluating the connection between national health expenditure and GDP. As already mentioned, attempts to detect other explanatory variables – or explanatory variables proper, since a correlation between health care expenditure and GDP does not explain much in terms of causal relations, unless one adopts a crude version of Keynesianism – have been sporadic and largely unsuccessful.³ Not even such obvious candidates as population shares above certain age thresholds (e.g., 65 years or 75 years) have been found to contribute to the explanation of health care expenditure – except in a few studies, e.g., Hitiris and Posnett (1992), Di Matteo and Di Matteo (1998), and Okunade et al. (2004). Zweifel et al. (1999) show that ‘proximity to death’ rather than ageing drives health care expenditure. Evidently, this variable is contemporaneously unknown and hence inoperative for models intending to forecast HCE.⁴

“Economists have not developed a formal theory to explain or to predict the per capita medical care expenditure of a nation”, concludes Wilson (1999, p. 160). He continues: “In the absence of a theory, empirical work in this field has necessarily been based on ad-hoc reasoning and data availability” (Wilson (1999, p. 160)).

Of course, Wilson's claim could be disputed on the grounds that Grossman (1972) has introduced a micro-founded model of the demand for health. But the Grossman model is concerned with the individual's demand for health, while Wilson bemoans the absence of a theory that would explain aggregate medical care expenditure. To reduce the Grossman model to a form that can be tested empirically with aggregate data has turned out to be cumbersome since important explanatory variables are unobservable (cf. Wagstaff, 1986, Erbsland et al., 1998, Nocera and Zweifel, 1998). After all, both Roberts (1999) and Gerdtham and Jönsson (2000) call for strengthening the theoretical basis for the macroeconomic analysis of health expenditure. According to Roberts (1999, p. 470), this must be the “primary aim for future work”.

We here intend to meet this demand, revisiting Baumol's (1967) theory of ‘unbalanced growth’. Baumol develops a simple neoclassical growth model that allows for explicit predictions of the future course of health expenditure. These predictions can be tested empirically. As far as we see, Baumol's model has so far largely escaped the attention of health economists.⁵ The next section introduces the model, and Section 3 discusses one of its more controversial implications – namely that labor productivity growth in the health sector must be expected to remain low – with special attention to measurement issues. Section 4 moves on to testing the main implications auf Baumol's model empirically. Section 5 presents the results of robustness checks; Section 6 draws some conclusions for health policy, and Section 7 concludes.