Baumol’s cost disease in acute versus long-term care: Do the differences loom large?

Celebi, Kaan; Hartwig, Jochen; Sandqvist, Anna Pauliina Article — Published Version Baumol’s cost disease in acute versus long-term care: Do the differences loom large? International Journal of Health Economics and Management Provided in Cooperation with: Springer Nature Suggested Citation: Celebi, Kaan; Hartwig, Jochen; Sandqvist, Anna Pauliina (2025) : Baumol’s cost disease in acute versus long-term care: Do the differences loom large?, International Journal of Health Economics and Management, ISSN 2199-9031, Springer US, New York, NY, Vol. 25, Iss. 2, pp. 159-191, https://doi.org/10.1007/s10754-025-09392-9 This Version is available at: https://hdl.handle.net/10419/330429 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ Vol.:(0123456789) International Journal of Health Economics and Management (2025) 25:159–191 https://doi.org/10.1007/s10754-025-09392-9 RESEARCH ARTICLE Baumol’s cost disease inacute versuslong‑term care: Do thedifferences loom large? KaanCelebi1· JochenHartwig1,2,3· AnnaPauliinaSandqvist4 Received: 14 April 2024 / Accepted: 9 January 2025 / Published online: 24 February 2025 © The Author(s) 2025 Abstract Baumol’s (Am Econ Rev 57: 415–426, 1967) model of ‘unbalanced growth’ yields a supply-side explanation for the ‘cost explosion’ in health care. Applying a testing strategy suggested by Hartwig (J Health Econ 27: 603–623, 2008), a sprawling literature affirms that the ‘Baumol effect’ has both a statistically and economically significant impact on health care expenditure growth. Skeptics maintain, however, that the proliferation of hi-tech medicine in acute care is clearly at odds with the assumption underlying Baumol’s model that productivity-enhancing machinery and equipment is only installed in the ‘progressive’ (i.e. manufacturing) sector of the economy. They argue that Baumol’s cost disease may affect long-term care, but not acute care. Our aim in this paper is to test whether Baumol’s cost disease affects long-term care and acute care differently. Our testing strategy consists in combining Extreme Bounds Analysis (EBA) with an outlier-robust MM estimator. Using panel data for 23 OECD countries, our results provide robust and statistically significant evidence that expenditures on both acute care and long-term care are driven by Baumol’s cost disease, even though the effect on long-term care expenditures is more pronounced. Keywords Health care expenditure· Baumol’s cost disease· Extreme Bounds Analysis· MM estimator· OECD panel JEL Classification C12· C23· I10 * Jochen Hartwig [email protected] 1 Faculty ofEconomics andBusiness Administration, Chemnitz University ofTechnology, Chemnitz, Germany 2 KOF Swiss Economic Institute, ETH Zurich, Zurich, Switzerland 3 Forum forMacroeconomics andMacroeconomic Policies, Hans Böckler Stiftung, Düsseldorf, Germany 4 Deloitte GmbH Wirtschaftsprüfungsgesellschaft, Munich, Germany 160 K.Celebi et al. Introduction Baumol’s (1967) model of ‘unbalanced growth’ yields a supply-side explanation for the ‘cost explosion’ in health care. Baumol divides the economy into two parts: a ‘progressive’ and a ‘non-progressive’ sector. He assumes that productivity growth is higher in the progressive (secondary) than in the non-progressive—or ‘stagnant’—(tertiary) sector of the economy, but wages grow more or less the same in both sectors. Therefore, unit costs and also prices rise much faster in the tertiary sector than in the secondary. Demand for certain services, like health care and education for instance, is hardly price-elastic, hence consumers are willing to pay the higher prices. Therefore, even if the two sectors keep their proportion in terms of real production, an ever-higher share of total expenditures will be channeled into the stagnant sector. This phenomenon is known as ‘Baumol’s cost disease’.1 Hartwig (2008) has suggested a test of whether Baumol’s cost disease drives health care expenditure (HCE) in OECD countries that does not require price or productivity data for the health sector, which are notoriously unreliable (see Berndt etal., 2000, p. 171). This test consists in regressing HCE growth rates (log differences) on the difference between nominal wage growth and labor productivity growth in the overall economy (plus controls). Rossen and Faroque (2016, p. 192) neatly summarize the intuition behind this approach as follows: “His [Hartwig’s] key insight is that since wage growth in health care depends on the higher productivity growth in the rest of the economy, growth in the unit labor cost and price of health care services, and therefore growth in health care spending, must bear a proportional relationship to the excess wage growth over labor productivity growth in the overall economy”.2 Evidently, Hartwig’s ‘Baumol variable’, i.e. the difference between nominal wage growth and labor productivity growth, equals the growth rate of aggregate (nominal) unit labor cost (NULC). To check whether the ‘Baumol variable’ is not just picking up purely monetary changes, Hartwig (2008) deflated both per-capita HCE— the dependent variable—and nominal wages per employee on the right-hand side of the regression equation by the GDP deflator. Hence, his ultimate test for whether Baumol’s cost disease drives health care expenditure consists in regressing real HCE growth on the growth rate of aggregate real unit labor cost (RULC) plus controls.3 Hartwig (2008) and scholars following his lead to testing Baumol’s cost disease in health care (see, e.g., Bates & Santerre, 2013, Medeiros & Schwierz, 2013, Hartwig & Sturm, 2014, Rossen & Faroque, 2016, Colombier, 2017, Tian etal., 2018, Bellido etal., 2019, Lorenzoni etal., 2019, Jeetoo, 2020, Wang & Chen, 2021) have thoroughly confirmed that the ‘Baumol effect’4 has both a statistically and economically significant impact on health care expenditure growth.5 1 The term ‘Baumol’s cost disease’ was coined by Alice Vandermeulen (1968), see Baumol (2012, p. xii). 2 See also Lorenzoni etal. (2019, p. 40) for a single-page derivation of Hartwig’s cost disease variable. 3 The growth rate of aggregate RULC equals the growth rate of the wage share in GDP. 4 Helland and Tabarrok (2019) prefer the expression ‘Baumol effect’ since it avoids the negative connotations of the term ‘Baumol’s (cost) disease’. 5 Rossen and Faroque (2016, p. 203), who only perform the nominal version of the test, conclude that “Baumol’s cost disease on health-care spending increases in Canada may not be economically very important”. Bates and Santerre (2013) and Wang and Chen (2021) also find relatively minor effects. The main reason for this finding, however, seems to be that these authors apply a ‘correction’ to Hartwig’s ‘Baumol variable’ suggested by Colombier (2012). The main effect of this ‘correction’ is that it scales the estimated coefficient down by a factor around 10 (see Table2 in Rossen and Faroque 2016). 161 Baumol’s cost disease inacute versuslong‑term care: Do the… Even though empirical research over the past decade and a half has built up a strong case in favor of Baumol’s cost disease being one of the main drivers of HCE growth, there remains one piece of skepticism that motivates our present paper. This skepticism was first brought to our attention by the late Gebhard Kirchgässner, then president of the Swiss federal Commission for Business Cycle Affairs (KfK). The 2006 annual report of that commission titled ‘Reforming the health system’ (Kommission für Konjunkturfragen, 2006, pp. 36–37) quotes the working paper version of Hartwig (2008) rather disapprovingly. Technological progress, so the argument goes, is rife in acute care, becoming manifest in hi-tech medicine. Therefore, the assumption underlying Baumol’s model of ‘unbalanced growth’ that productivity-enhancing machinery and equipment is only installed in the secondary sector of the economy is clearly flawed. The report does concede an impact of Baumol’s cost disease on HCE growth, but restricts it to the long-term care (LTC) sector (see also Kirchgässner, 2009). Similarly, de la Maisonneuve and Oliveira Martins (2014), in their health expenditure projections until 2060 on behalf of the OECD, model (public) HCE and LTC expenditure separately and allow Baumol’s cost disease (for which they use the level or the growth rate of labor productivity in the total economy as a proxy) only to affect the latter. In their update of the OECD’s spending projections, however, Lorenzoni etal. (2019, p. 25), note that the allowance of “the impact of the Baumol effect on health care as a whole (instead of only for long-term care)” was one of the main differences against previous studies. Our aim in this paper is to test whether Baumol’s cost disease affects long-term care and acute care differently. Acute care, according to the OECD (2008, p. 17) “is one in which the principal intent is one or more of the following: (i) to manage labour (obstetrics), (ii) to cure illness or to provide definitive treatment of injury, (iii) to perform surgery, (iv) to relieve symptoms of illness or injury (excluding palliative care), (v) to reduce severity of an illness or injury, (vi) to protect against exacerbation and/or complication of an illness and/or injury which could threaten life or normal function, (vii) to perform diagnostic or therapeutic procedures”. In the Wikipedia entry on ‘acute care’ it reads: “In medical terms, care for acute health conditions is the opposite from chronic care, or longer-term care”.6 We, therefore, define acute care expenditure (ACE) very broadly as total current health care expenditure (HCE) minus long-term care expenditure (LTCE). Acute care is provided as inpatient or outpatient care by hospitals and medical and dental practices or other healthcare providers. Long-term care, on the other hand, comprises (i) medical or nursing care, such as relieving pain and other symptoms, (ii) personal care services, i.e. help with activities of daily living, performed either by relatives or nursing staff and (iii) assistance services, which enable persons to live independently at home, e.g. shopping or performing housework (OECD etal., 2017, p. 91). The modes of provision of long-term care are (i) inpatient longterm care in hospitals or nursing homes requiring an overnight stay with medical supervision, (ii) day cases of long-term care delivered by the same providers, but without an overnight stay and (iii) outpatient or home-based long-term care, which typically involves providers of nursing services regularly visiting elderly people who are becoming more dependent (OECD etal., 2017, pp. 94–95). In order to test whether Baumol’s cost disease affects long-term care and acute care differently we use the same strategy as Hartwig and Sturm (2014), i.e. Extreme Bounds Analysis (EBA) combined with an outlier-robust MM estimator. As EBA includes many (if 6 See https:// en. wikip edia. org/ wiki/ Acute_ care. 162 K.Celebi et al. not all) of the HCE drivers that have been suggested in the literature, and omitted variables are an important source of endogeneity, considering many explanatory variables mitigates the latter. However, we do not claim to properly identify causal effects. When we use the term ‘effect’ in our empirical analysis and often when we refer to the literature, it relates to conditional correlations, not a causal relationship. In other words, we are suggesting rather than testing for causal relationships. The remainder of this paper is structured as follows. The next section discusses our dataset. Sect. "Methodology" explains the methodologies of Extreme Bounds Analysis and outlier-robust MM-estimation. Sects. "Results" and "Robustness tests" present the results—including those of robustness checks—and Sect."Conclusion" concludes. Data issues around LTCE are discussed in Appendix1. Data The data source for most of the variables is the OECD Health Database, which also contains economic, socio-demographic, and even technological data (as long as they are health-related).7 Considering the dependent variables, data on total health expenditures are available for quite a long time period and many countries while data on long-term care expenditures (LTCE) tend to be relatively scarce. We exclude Ireland, Italy, New Zealand, and the United Kingdom (because of the small number of observations) and countries with a very low share (smaller than 5 percent in 2017) of LTCE in HCE (Australia, Greece, Hungary, Latvia, Portugal, and Slovakia) as their data might be of low quality and/or noisy. Our dataset thus covers the following 23 OECD countries: Austria, Belgium, Canada, the Czech Republic, Denmark, Estonia, Finland, France, Germany, Iceland, Israel, Japan, South Korea, Lithuania, Luxembourg, the Netherlands, Norway, Poland, Slovenia, Spain, Sweden, Switzerland, and the United States. With respect to the explanatory variables, numerous possible determinants have been introduced. We draw on Hartwig and Sturm (2014), who have conducted an extensive literature review to uncover all macroeconomic and institutional determinants of HCE growth that have been suggested in the literature, and introduce them in an EBA framework to be explained in the next section.8 Most of these variables are also drawn from the OECD Health Database. Against Hartwig and Sturm (2014), we updated the sample as far as possible and included sugar intake, the importance of which as HCE driver was demonstrated by Castro (2017).9 Our sample period runs from 1971 to 2019. Variable descriptions and summary statistics are given in Tables1 and 2. 7 We mostly used the 2020 version of the OECD Health database. All data are available on request to the corresponding author. 8 Hartwig and Sturm’s data were also used by Hauck and Zhang (2016). 9 We found no sources to update the dummy variables used by Hartwig and Sturm (2014) to model the institutional specifics of the national health systems. Since none of these institutional dummy variables turned out as a robust and statistical significant explanatory variable for HCE growth in Hartwig and Sturm’s EBAs, and in order not to forego the most recent years by including the non-updated dummies, we decided to drop them from our analysis. 163 Baumol’s cost disease inacute versuslong‑term care: Do the… Methodology Baseline results To produce baseline results we regress the growth rate of HCE/ACE/LTCE in real terms on the growth rate of real GDP per capita and on the Baumol variable (i.e. the growth rate of the wage share). We include real GDP because of the longstanding insight originating from Newhouse (1977) that GDP (or income) drives HCE. Research into the determinants of HCE growth since Newhouse’s pioneering study has for a long time failed to disclose other robust explanatory variables beyond national income growth (see Roberts, 1999). Therefore, we include only GDP growth and the Baumol variable in our baseline model as well as in the M-vector of the Extreme Bounds Analysis to be discussed below. The models are estimated with OLS using country-clustered robust standard errors as well as with an outlier-robust MM estimator. The following specifications are estimated: the first model includes only a constant, the second one additionally fixed country effects (FE country) and the third model fixed year effects (FE time). The fourth specification includes both fixed country and year effects (FE both). Furthermore, a fifth OLS specification incorporating country-specific trends (CST) is included.10 Extreme bounds analysis To examine the sensitivity of the individual variables on per-capita HCE growth, we apply (variants of) EBA, as suggested by Leamer (1985) and Levine and Renelt (1992).11 This approach, which has been widely used in the economic growth literature, has become a popular tool for economists who want to test the robustness of the results of their empirical work. In addition, the EBA provides an opportunity to test whether a particular determinant is robustly related to the dependent variable. The central difficulty in this research—which also applies to the research topic of the present paper—is that several different models may all seem reasonable given the data but yield different conclusions about the parameters of interest. As argued by Temple (2000), it is rare in empirical research that we can say with certainty that one model dominates all other possibilities in all dimensions. In these circumstances, it makes sense to provide information about how sensitive the findings are to alternative modelling choices. EBA provides a relatively simple means of doing exactly this. It involves systematically testing all possible combinations of variables in a regression model. Specifically, the EBA involves running a large number of regressions, each with a different combination of variables to see how sensitive the estimated coefficients are to changes in the specification. For each regression, the coefficient of interest and the associated t-statistic are recorded. Finally, the distribution of these coefficients and t-statistics across all the regressions is examined to determine whether the coefficient of interest is robust to changes in the specification. Equations of the following general form are estimated: where Y is the dependent variable; M is a vector of ‘standard’ explanatory variables that will be included in each regression model; F is the variable of interest; Z is a vector of (1) Y = 𝛼M + 𝛽F + 𝛾Z + u, 10 For technical reasons, the function lmrob cannot calculate MM estimates with country-specific trends. 11 Parts of this section rely upon previous work (Hartwig and Sturm 2014). 164 K.Celebi et al. Table 1 Variable descriptions Variable code Variable label Description Source Transformation Baumol Baumol Compensation of employees as percentage of gross value added OECD Difference of log gdp GDP p.c GDP per capita in US-dollars PPP OECD Difference of log pop65 Population ≥ 65years Share of population 65years and over OECD/Eurostat First difference pop80 Population ≥ 80years Share of population 80years and over OECD/Eurostat First difference frp1564 Female p.r Female participation in the labor force (% of active pop.) OECD First difference ur Unemployment rate Unemployed as a share of the labor force (%) OECD First difference ta Health administration spending Per capita real expenditure on health administration, governance and health system and financing administration, per capita, constant prices, OECD base year OECD Difference of log accident Road fatalities Land traffic accidents, deaths per 100,000 population OECD Difference of log alcc Alcohol consumption Alcohol intake, liters per capita 15 + OECD Difference of log dp Population density Population density per square kilometer OECD First difference LE65.F Female life expectancy (65) Life expectancy at age 65 for females OECD Difference of log LE65.M Male life expectancy (65) Life expectancy at age 65 for males OECD Difference of log mort Mortality rate (0–69) Mortality rate, potential years of life lost per 100 000 population, 0–69 OECD First difference tobc Tobacco consumption Tobacco consumption, grams per capita per year 15 + OECD Difference of log covero Insurance coverage Insurance coverage % of total population covered OECD First difference dl.gerd Health R&D Gross expenditure on R&D, compound annual growth rate OECD – sugar Sugar supply Sugar supply in kilograms per capita per year OECD Difference of log bedsi Curative beds (per 1,000) Curative care beds per 1000 inhabitants OECD Difference of log bedsh Curative beds (per hospital) Curative care beds per general hospital OECD Difference of log gsh Public expenditure Public expenditure as percentage of GDP OECD One year lagged first difference puhes Public-to-health expenditure ratio Public expenditure as a share of total health expenditure OECD First difference texmc Inpatient expenditure Share of inpatient expenditure in total health expenditure OECD First difference hpi Health priceindex Price index for total expenditure on health Eurostat Difference of log doctca Physicians The stock of practicing physicians per 1000 population OECD Difference of log 165 Baumol’s cost disease inacute versuslong‑term care: Do the… Table 1 (continued) Variable code Variable label Description Source Transformation nurca Nurses Number of actively employed nurses per 1000 population OECD Difference of log persh Hospital employment Total hospital employment OECD Difference of log physh Physicians per 100 beds Physicians per 100 hospital beds OECD Difference of log rat Specialist-to-GP ratio The ratio of specialist to general practitioners OECD First difference rend Renal dialysis Renal dialysis rate per million inhabitants OECD Difference of log 166 K.Celebi et al. up to three possible additional explanatory variables, which the literature suggests may be related to the dependent variable; and u is an error term. The extreme bounds test for variable F states that if the lower extreme bound for β—the lowest value for β minus two standard deviations—is negative, and the upper extreme bound for β—the highest value for β plus two standard deviations—is positive, the variable F is not robustly related to Y. A main limitation of the method of EBA is that it cannot decently cope with strong multicollinearity. Two highly correlated variables often turn individually insignificant when entered jointly and should therefore ideally not both enter the EBA. Given the large Table 2 Summary statistics *indicates difference of log, which is represented with the prefix “dl_” throughout the paper. ** indicates first difference, which is represented with the prefix “d_” throughout the paper Variable label Observations Mean SD Min Max Total health care expenditure* 581 0.5 4.2 − 27.8 23.0 Long-term care expenditure* 581 5.2 25.1 − 24.8 487.2 Acute care expenditure* 581 0.2 3.9 − 28.5 12.7 Baumol* 580 0.0 2.4 − 13.5 13.2 GDP p.c.* 581 1.8 2.7 − 15.4 11.7 Population ≥ 65years** 581 0.2 0.2 − 0.4 1.0 Population ≥ 80years** 579 0.1 0.1 − 0.4 0.7 Female p.r.** 574 0.5 0.8 − 2.6 7.7 Unemployment rate** 480 − 0.1 1.2 − 4.4 8.1 Health administration spending* 572 3.1 17.0 − 99.4 166.5 Road fatalities* 580 − 4.6 15.2 − 133.3 137.6 Alcohol consumption* 576 − 0.2 4.3 − 27.1 26.2 Population density** 480 1.0 1.6 − 4.5 8.1 Female life expectancy (65)* 575 0.7 1.4 − 5.7 8.1 Male life expectancy (65)* 575 0.9 1.4 − 6.7 8.1 Mortality rate (0–69)* 531 − 1.1 21.8 − 127.9 94.1 Tobacco consumption* 398 − 2.5 10.5 − 80.8 86.0 Insurance coverage** 520 0.1 1.7 − 5.6 36.4 Health R&D* 486 4.2 6.3 − 17.7 56.7 Sugar supply* 555 0.9 9.3 − 37.9 115.8 Curative beds (per 1,000)* 419 − 1.3 7.1 -53.2 103.2 Curative beds (per hospital)* 419 − 0.7 8.5 − 93.2 53.9 Public expenditure** 575 0.1 0.4 − 1.0 5.9 Public-to-health expenditure ratio** 581 0.1 1.9 − 7.8 35.2 Inpatient expenditure** 580 − 0.1 1.9 − 35.2 7.8 Health price index* 365 2.3 3.0 − 7.9 27.0 Physicians* 435 1.5 2.5 − 14.7 29.4 Nurses* 369 1.7 2.9 − 9.0 20.7 Hospital employment* 304 1.0 14.1 − 94.4 131.7 Physicians per 100 beds** 329 − 0.3 12.4 − 171.7 18.2 Specialist-to-GP ratio** 406 0.0 0.5 − 2.2 4.8 Renal dialysis* 410 2.5 9.7 − 116.8 36.4 173 Baumol’s cost disease inacute versuslong‑term care: Do the… ACE and LTCE in both OLS and MM estimations is significant at the 1% level.20 This suggests that the Baumol coefficient is significantly higher for LTCE as dependent variable than for ACE.21 The importance of income (GDP) seems to be lower in explaining long-term care expenditure compared with acute care or overall health expenditure. GDP growth remains robust, but the coefficient is lower than in Tables8 and 9, and the variable is statistically significant in less than 50% of the regressions in the MM-EBA (and less than 10% in the OLS-EBA). Other robust variables ‘outperform’ GDP growth in terms of higher proportions of significant coefficients, for instance, the (change in the) unemployment rate (d.ur),22 the change in the share of the population 80years and over (d.pop80) and the growth rate of the stock of practicing physicians per 1000 population (dl.doctca).23 One reason why we think the MM-EBA results are more plausible than the OLS-EBA results for LTCE is that the coefficient on d.pop80 is negative on average in the OLS estimations. Table 6 Tests—Baumol coefficient LTCE versus ACE This table presents the test statistics and corresponding p values of a Wald test which assesses whether the estimated coefficient of the Baumol variable is equal in the ACE and LTCE estimates OLS MM Constant FE country FE time FE both CST Constant FE country FE time FE both Test-statistic 1.923 0.032 0.517 0.012 0.009 − 0.567 0.236 0.564 0.833 P value 0.166 0.859 0.472 0.914 0.924 0.285 0.593 0.713 0.797 Table 7 Tests—GDP coefficient LTCE versus ACE This table presents the test statistics and corresponding p values of a Wald test which assesses whether the estimated coefficient of GDP is equal in the ACE and LTCE estimates OLS MM Constant FE country FE time FE both CST Constant FE country FE time FE both Test-statistic 2.951 1.577 2.497 0.952 0.005 − 1.681 − 1.145 − 1.440 − 1.269 P value 0.086 0.209 0.114 0.329 0.945 0.047 0.126 0.075 0.103 20 In the case of the Baumol coefficients estimated with OLS, we obtain: (4) t =𝜇ACE − 𝜇LTCE √ 𝜎2 ACE n ACE + 𝜎2 LTCE n LTCE = 0.642 − 0.787 √ 0.05002 20853 +0.25592 20853 =− 80.39 , whereas for the Baumol coefficients estimated with MM we get: (5) t = 0.693−0.923 √ 0.04072 1179 +0.11262 1972 =− 81.92 . 21 This result puts into perspective our earlier finding based on a seemingly unrelated regression that the impact of the Baumol variable on ACE is not different from its impact on LTCE (see Table6). 22 The impact of unemployment on LTCE growth is much higher than on ACE growth. 23 That the coefficient on dl.doctca is negative on average may seem implausible given that ‘too many doctors’ are thought to increase health costs. This seems not to be the case in long-term care, though. 174 K.Celebi et al. Table 8 EBA results for HCE (with year and country FE) OLS MM Variables Mean βMin βMax βØ se %Sign CDF(0) Mean βMin βMax βØ se %Sign CDF(0) dl.Baumol 0.69 0.53 0.84 0.07 100.0 100.0 0.72 0.60 0.85 0.08 100.0 100.0 dl.gdp 0.61 0.34 1.18 0.10 100.0 100.0 0.59 0.38 1.26 0.10 100.0 100.0 d.ur 0.40 0.04 1.16 0.21 37.5 100.0 0.48 0.28 0.99 0.19 76.6 100.0 dl.hpi − 0.15 − 0.32 0.02 0.06 63.1 99.1 0.00 − 0.15 0.12 0.07 0.6 57.0 dl.tobc 0.01 0.00 0.02 0.01 0.0 98.1 0.01 0.00 0.02 0.01 0.7 98.7 d.gsh_l1 0.60 − 0.26 2.78 0.46 12.7 96.0 0.61 0.14 2.18 0.32 25.5 100.0 d.physh 0.02 − 0.04 0.12 0.02 33.8 93.0 0.03 − 0.06 0.10 0.02 30.2 93.4 dl.sugar 0.02 − 0.02 0.04 0.02 0.0 91.1 0.00 − 0.01 0.02 0.01 0.0 74.4 dl.bedsh 0.01 − 0.16 0.06 0.02 1.6 86.4 0.00 − 0.04 0.03 0.02 0.0 79.2 dl.ta 0.01 − 0.02 0.08 0.01 22.5 85.7 0.01 − 0.01 0.06 0.01 13.2 95.6 d.puhes 0.28 -2.53 16.75 0.24 3.7 83.2 0.07 − 2.31 7.69 0.08 3.0 56.8 d.rat − 0.14 − 0.81 0.52 0.37 0.0 82.8 − 0.27 − 0.59 0.18 0.35 3.4 98.6 dl.persh 0.01 − 0.04 0.10 0.02 13.6 82.2 0.02 − 0.04 0.07 0.02 17.9 85.5 dl.accident 0.00 − 0.03 0.01 0.01 2.1 81.9 − 0.01 − 0.02 0.00 0.01 0.8 100.0 d.texmc 0.20 − 2.54 16.71 0.24 3.0 81.8 0.06 − 2.33 7.69 0.08 3.0 53.8 d.pop80 1.03 − 3.15 6.66 2.16 0.5 80.9 0.09 − 3.64 2.31 1.52 0.0 53.2 dl.LE65.M − 0.10 − 0.37 0.40 0.13 5.3 80.6 − 0.03 − 0.19 0.19 0.12 0.0 60.2 dl.gerd − 0.01 − 0.07 0.07 0.03 0.3 79.1 − 0.01 − 0.04 0.06 0.03 0.0 83.2 dl.nurca 0.02 − 0.08 0.14 0.07 0.0 78.6 0.02 − 0.03 0.15 0.04 1.6 83.3 d.frp1564 0.14 − 0.29 0.68 0.24 0.0 78.4 − 0.08 − 0.30 0.11 0.21 0.0 66.4 d.mort 0.01 − 0.02 0.04 0.01 2.6 78.1 0.00 0.00 0.02 0.01 0.0 96.4 dl.doctca − 0.05 − 0.29 0.22 0.09 11.0 74.6 − 0.08 − 0.23 0.20 0.09 1.1 97.8 dl.bedsi 0.01 − 0.11 0.11 0.03 0.7 73.5 0.03 − 0.03 0.07 0.03 8.6 97.1 d.dp 0.07 − 0.36 0.61 0.26 0.0 69.7 0.01 − 0.24 0.48 0.17 0.5 52.2 dl.LE65.F − 0.08 − 0.55 0.47 0.16 6.5 67.0 − 0.02 − 0.37 0.27 0.13 0.8 62.0 175 Baumol’s cost disease inacute versuslong‑term care: Do the… This table shows the results of OLS-EBA and MM-EBA estimations using HCE growth as the dependent variable. For the explanatory variables, the average estimated coefficient (mean β), minimum and maximum of the coefficients (min β and max β), their average standard deviation (Ø se), the proportion of significant coefficients at the 5% level (%Sign.) and the cumulative distribution function CDF(0) of the estimated coefficients are reported. CDF(0) values > 90% and %Sign values > 90% are highlighted in bold Table 8 (continued) OLS MM Variables Mean βMin βMax βØ se %Sign CDF(0) Mean βMin βMax βØ se %Sign CDF(0) dl.alcc 0.02 − 0.04 0.28 0.04 3.5 65.2 0.01 − 0.01 0.08 0.03 0.0 71.3 dl.rend − 0.01 − 0.05 0.04 0.02 5.4 62.3 0.00 − 0.03 0.05 0.02 0.0 60.7 d.covero − 0.01 − 0.49 2.72 0.23 0.0 62.0 0.14 − 0.19 2.25 0.17 7.3 66.0 d.pop65 0.18 − 2.87 2.59 1.08 0.1 57.6 0.04 − 0.74 0.93 0.75 0.0 51.4 176 K.Celebi et al. Table 9 EBA results for ACE (with year and country FE) OLS MM Variables Mean βMin βMax βØ se %Sign CDF(0) Mean βMin βMax βØ se %Sign CDF(0) dl.Baumol 0.64 0.50 0.84 0.07 100.0 100.0 0.69 0.55 0.85 0.08 100.0 100.0 dl.gdp 0.62 0.34 1.17 0.09 100.0 100.0 0.57 0.34 1.20 0.10 100.0 100.0 d.ur 0.39 0.05 1.27 0.20 48.5 100.0 0.49 0.30 0.98 0.22 67.4 100.0 dl.tobc 0.01 0.00 0.03 0.01 0.0 100.0 0.01 0.00 0.02 0.01 1.4 100.0 dl.hpi − 0.15 − 0.31 0.07 0.06 69.5 97.1 − 0.10 − 0.29 0.14 0.12 9.8 93.0 dl.accident − 0.01 − 0.04 0.00 0.01 11.3 97.0 − 0.01 − 0.04 − 0.01 0.01 28.2 100.0 d.gsh_l1 0.64 − 0.30 2.87 0.43 18.8 96.7 0.42 0.08 1.73 0.27 8.7 100.0 dl.persh 0.02 − 0.04 0.08 0.01 10.5 90.5 0.01 − 0.02 0.07 0.02 12.2 82.7 dl.sugar 0.01 − 0.02 0.05 0.02 0.0 85.7 0.00 − 0.03 0.01 0.01 0.0 53.7 dl.gerd − 0.02 − 0.10 0.06 0.03 3.8 82.8 − 0.01 − 0.05 0.04 0.03 1.6 81.3 dl.alcc 0.02 − 0.05 0.21 0.03 6.8 82.0 0.01 − 0.05 0.08 0.03 0.0 70.5 d.physh 0.01 − 0.07 0.10 0.02 4.0 81.5 0.02 − 0.10 0.10 0.02 15.0 90.0 dl.nurca 0.02 − 0.08 0.10 0.07 0.0 77.3 0.02 − 0.03 0.13 0.05 0.0 68.0 dl.ta 0.01 − 0.02 0.12 0.01 14.1 74.1 0.02 − 0.01 0.15 0.01 52.3 99.2 d.rat − 0.09 − 0.79 0.83 0.35 0.1 72.3 − 0.33 − 0.77 0.23 0.41 2.8 98.6 dl.LE65.M − 0.06 − 0.36 0.34 0.12 0.3 69.0 − 0.03 − 0.19 0.15 0.13 0.0 74.0 dl.bedsi 0.01 − 0.07 0.15 0.03 0.1 69.0 0.01 − 0.04 0.05 0.02 0.0 91.5 dl.bedsh 0.00 − 0.14 0.06 0.02 1.0 68.7 0.00 − 0.05 0.02 0.01 0.5 70.5 d.dp 0.04 − 0.30 0.59 0.23 0.0 62.4 − 0.04 − 0.23 0.33 0.20 0.0 71.3 d.frp1564 − 0.05 − 0.43 0.42 0.22 0.0 62.3 − 0.10 − 0.30 0.19 0.23 0.0 84.7 dl.LE65.F 0.00 − 0.45 0.45 0.15 0.4 56.9 0.01 − 0.29 0.35 0.13 0.0 59.5 d.covero 0.01 − 0.46 3.16 0.21 0.0 56.6 0.09 − 0.27 1.40 0.20 0.0 55.2 d.texmc 0.15 − 4.06 18.49 0.23 2.3 55.1 0.33 − 1.87 7.74 0.15 5.6 62.6 d.mort 0.00 − 0.02 0.03 0.01 0.0 54.4 0.00 − 0.01 0.02 0.01 0.0 88.3 d.pop65 − 0.11 − 2.23 2.37 1.00 0.0 53.7 0.08 − 0.61 1.40 0.81 0.0 60.2 177 Baumol’s cost disease inacute versuslong‑term care: Do the… This table shows the results of OLS-EBA and MM-EBA estimations using ACE growth as the dependent variable. For the explanatory variables, the average estimated coefficient (mean β), minimum and maximum of the coefficients (min β and max β), their average standard deviation (Ø se), the proportion of significant coefficients at the 5% level (%Sign.) and the cumulative distribution function CDF(0) of the estimated coefficients are reported. CDF(0) values > 90% and %Sign values > 90% are highlighted in bold Table 9 (continued) OLS MM Variables Mean βMin βMax βØ se %Sign CDF(0) Mean βMin βMax βØ se %Sign CDF(0) dl.doctca − 0.02 − 0.29 0.30 0.08 7.4 53.7 − 0.05 − 0.24 0.22 0.08 2.3 86.4 dl.rend 0.00 − 0.04 0.06 0.02 0.1 53.3 0.00 − 0.02 0.03 0.02 0.0 67.4 d.puhes 0.17 − 4.08 18.52 0.23 2.4 52.3 0.33 − 1.86 7.73 0.14 5.4 59.5 d.pop80 − 0.05 − 4.71 3.79 2.00 0.0 50.7 − 0.69 − 4.47 1.29 1.41 0.0 82.5 178 K.Celebi et al. Table 10 EBA results for LTCE (with year and country FE) OLS MM Variables Mean βMin βMax βØ se %Sign CDF(0) Mean βMin βMax βØ se %Sign CDF(0) dl.tobc − 0.04 − 0.09 0.00 0.07 0.0 100.0 0.00 − 0.02 0.02 0.02 0.0 87.0 dl.Baumol 0.79 − 0.29 1.59 0.51 49.5 99.7 0.92 0.52 1.22 0.13 99.9 100.0 dl.LE65.M − 1.27 − 3.93 0.86 0.96 9.2 98.0 − 0.05 − 0.31 0.25 0.18 0.0 57.7 dl.gdp 0.88 − 0.86 3.45 0.69 7.7 97.7 0.41 0.10 1.27 0.21 43.5 100.0 d.mort 0.12 − 0.09 0.42 0.08 24.3 95.7 0.01 − 0.02 0.03 0.01 0.0 84.5 dl.gerd 0.38 − 0.12 0.96 0.23 32.7 95.1 − 0.03 − 0.09 0.20 0.05 2.4 82.0 dl.sugar 0.11 − 0.19 0.23 0.14 0.0 95.1 0.02 − 0.01 0.04 0.02 0.0 92.3 dl.ta 0.12 − 0.16 0.51 0.07 39.6 94.1 − 0.02 − 0.14 0.03 0.02 9.7 71.3 d.texmc − 0.56 − 14.87 27.99 1.69 15.5 91.0 0.20 − 3.87 15.34 0.25 15.9 64.0 d.physh 0.08 − 0.20 0.85 0.10 0.3 90.9 0.02 − 0.42 0.23 0.04 27.8 50.4 d.puhes 1.07 − 13.72 30.23 1.69 14.8 90.3 0.21 − 3.86 15.36 0.24 18.6 54.3 dl.alcc 0.31 − 0.51 1.03 0.27 0.4 88.3 − 0.02 − 0.19 0.20 0.08 0.5 75.1 d.ur 0.98 − 1.56 3.40 1.67 0.6 87.1 0.81 0.40 1.17 0.32 87.1 100.0 d.pop65 9.82 − 6.69 31.89 8.01 19.1 86.9 − 1.24 − 2.83 0.77 1.29 1.9 94.7 dl.persh − 0.08 − 0.36 0.24 0.09 2.1 85.1 0.01 − 0.04 0.13 0.03 7.1 69.5 d.frp1564 1.41 − 2.62 6.36 1.75 25.1 81.9 0.06 − 0.32 0.65 0.30 0.0 56.9 dl.LE65.F − 0.43 − 2.24 5.63 1.19 3.9 79.6 − 0.21 − 0.74 0.01 0.21 7.8 99.1 d.dp 0.69 − 2.29 3.14 1.98 0.0 79.6 0.54 − 0.04 1.30 0.30 38.6 99.6 dl.nurca − 0.29 − 1.28 0.63 0.58 0.0 76.4 0.02 − 0.05 0.25 0.06 6.1 58.2 dl.accident 0.02 − 0.09 0.11 0.07 0.0 75.5 0.02 -0.01 0.03 0.01 2.7 99.1 dl.doctca − 1.38 − 7.10 0.86 0.67 56.8 69.6 − 0.18 − 0.40 0.11 0.10 64.1 99.5 dl.bedsi − 0.14 − 2.19 0.56 0.28 2.4 67.7 − 0.04 − 0.26 0.59 0.06 5.4 98.2 dl.rend − 0.03 − 0.21 0.10 0.10 4.4 66.9 − 0.02 − 0.15 0.04 0.03 4.2 82.1 d.pop80 − 15.49 − 85.83 28.79 16.03 5.7 65.0 6.57 1.10 11.00 3.06 68.6 100.0 d.gsh_l1 − 0.98 − 10.95 3.09 3.43 0.0 63.5 0.64 − 0.22 1.73 0.42 20.4 98.1 179 Baumol’s cost disease inacute versuslong‑term care: Do the… This table shows the results of OLS-EBA and MM-EBA estimations using LTCE growth as the dependent variable. For the explanatory variables, the average estimated coefficient (mean β), minimum and maximum of the coefficients (min β and max β), their average standard deviation (Ø se), the proportion of significant coefficients at the 5% level (%Sign.) and the cumulative distribution function CDF(0) of the estimated coefficients are reported. CDF(0) values > 90% and %Sign values > 90% are highlighted in bold Table 10 (continued) OLS MM Variables Mean βMin βMax βØ se %Sign CDF(0) Mean βMin βMax βØ se %Sign CDF(0) dl.bedsh − 0.16 − 1.14 0.44 0.18 7.1 63.3 0.03 − 0.06 0.13 0.03 7.9 98.7 dl.hpi 0.39 − 1.16 1.57 0.52 17.0 53.1 − 0.14 − 0.72 0.17 0.19 4.7 91.8 d.covero − 0.02 − 6.87 3.04 1.22 2.7 51.9 − 0.04 − 0.50 0.71 0.27 2.1 54.2 d.rat − 0.28 − 4.23 5.83 3.02 0.0 50.7 − 0.60 − 1.97 0.48 0.81 8.1 89.2 180 K.Celebi et al. Robustness tests Figure1 shows the results of the jackknife test for HCE. The lowest estimated coefficients for the GDP and Baumol variables are 0.50 and 0.64, respectively. These were obtained when South Korea and Norway (respectively) were excluded. The highest estimated coefficient for GDP is found by excluding Canada (0.56). For the Baumol variable, the highest coefficient (0.71) results when excluding Spain. The mean values of the 23 coefficients calculated by the jackknife test are 𝜇GDP = 0.53 and 𝜇Baumol = 0.69 . With a standard deviation of 0.0144 and 0.0138 respectively, the estimated coefficients on GDP and the Baumol variable remained relatively robust throughout this experiment. It is also striking that in all iterations the estimated coefficients for both independent variables are significant at the 1% level. Results of the jackknife test are very similar when ACE instead of HCE is the dependent variable (see Fig.2). The jackknife results for LTCE as dependent variable are displayed in Fig. 3. The mean values and standard deviations of the coefficients are: 𝜇GDP = 0.24 , 𝜎GDP = 0.0475, 𝜇Baumol = 0.74 , 𝜎Baumol = 0.0203 .24 Excluding Iceland results in the lowest GDP coefficient (0.10), while excluding Estonia produces the highest one (0.31). The lowest Baumol coefficient (0.71) is obtained by excluding Iceland, while the highest one (0.78) is obtained by excluding Sweden. All estimated Baumol coefficients remain statistically significant at the 1% level; GDP growth, however, does not contribute to explaining LCTE growth at conventional levels of significance. This confirms our finding from Table5 for the outlier-robust MM-estimation with country and time fixed effects. The results of the EBA estimations performed with lagged independent variables are shown in Tables 11, 12 and 13 in Appendix 2. When HCE or ACE are considered as dependent variables, high CDF(0) values indicate that the Baumol variable remains a robust explanatory variable, although its lagged version is statistically significant in fewer regressions than its un-lagged counterpart, and the coefficient estimates are considerably lower. Lagged GDP growth is also a quite robust explanatory variable for both HCE and ACE growth, with relatively high CDF(0) values. This is not the case for LTCE growth, however. In the LTCE EBA estimated with OLS, the lagged Baumol variable drops below the 90% CDF(0) threshold. This is due to outliers, however, as the MM estimation demonstrates. Even when lagged by one year, the Baumol variable remains a robust explanatory variable for LTCE growth. Conclusion Baumol’s cost disease has been argued to be a major driver of health care expenditure growth. Skeptics doubt its impact on expenditures on acute care, however, maintaining that only labor intensive tasks in long-term care are likely to be affected by the cost disease. Our aim in this paper was to test this proposition empirically. In our analysis, we combine Extreme Bounds Analysis (EBA) with outlier-robust MM estimation and use data for 23 OECD countries over the period 1971–2019. We find that, although the impact of Baumol’s cost disease is strongest on long-term care expenditure, 24 Revisiting the question whether the Baumol effect affects ACE and LTCE differently, these findings reinforce the conclusion that the effect on LTCE is stronger. 181 Baumol’s cost disease inacute versuslong‑term care: Do the… Fig. 1 Jackknife results for HCE Note: The dependent variable is HCE growth. The figure shows the estimated coefficients and their corresponding p values for GDP growth and Baumol in each iteration, omitting one of the 23 countries in each step. The coefficients were estimated using outlier-robust MM-estimation with country and time fixed effects, using clustered standard errors to calculate the p values. The mean 𝜇 and standard deviation 𝜎 of the 23 coefficients calculated by the jackknife test are 𝜇GDP = 0.53 , 𝜇Baumol = 0.69 , 𝜎GDP = 0.0144 and 𝜎Baumol = 0.0138 Fig. 2 Jackknife results for ACE Note: The dependent variable is ACE growth. The figure shows the estimated coefficients and their corresponding p values for GDP growth and Baumol in each iteration, omitting one of the 23 countries in each step. The coefficients were estimated using outlier-robust MM-estimation with country and time fixed effects, using clustered standard errors to calculate the p values. The mean 𝜇 and standard deviation 𝜎 of the 23 coefficients calculated by the jackknife test are 𝜇GDP = 0.54 , 𝜇Baumol = 0.68 , 𝜎GDP = 0.0129 and 𝜎Baumol = 0.0159 182 K.Celebi et al. expenditure on acute care is nevertheless also affected. These findings are robust to excluding single countries from the sample as well as to lagging all explanatory variables by one year. We conclude that Baumol’s cost disease drives the whole range of health care expenditures, not just those on labor intensive care work. Our results hence give succor to the decision taken by the OECD in 2019 to revise the methodology used for the organization’s health spending projections to allow for an “impact of the Baumol effect on health care as a whole (instead of only for long-term care)” (Lorenzoni etal., 2019, p. 25). Appendix1: Data issues aroundLTCE As OECD etal. (2017, p. 88) note, “(t)o date, estimates of spending on services for longterm health care have been mostly limited to higher-income countries, due to the fact that in most lowand middle-income countries (LMIC) long-term health care is provided as Fig. 3 Jackknife results for LTCE Note: The dependent variable is LTCE growth. The figure shows the estimated coefficients and their corresponding p values for GDP growth and Baumol in each iteration, omitting one of the 23 countries in each step. The coefficients were estimated using outlier-robust MM-estimation with country and time fixed effects, using clustered standard errors to calculate the p values. The mean 𝜇 and standard deviation 𝜎 of the 23 coefficients calculated by the jackknife test are 𝜇GDP = 0.24 , 𝜇Baumol = 0.74 , 𝜎GDP = 0.0475 and 𝜎Baumol = 0.0203 189 Baumol’s cost disease inacute versuslong‑term care: Do the… OLS-EBA and MM-EBA estimations, with LTCE growth as the dependent variable and explanatory variables lagged by one year. For the explanatory variables, the average estimated coefficient (mean β), minimum and maximum of the coefficients (min β and max β), their average standard deviation (Ø se), the proportion of significant coefficients (%Sign.) and the cumulative distribution function CDF(0) of the estimated coefficients are reported.CDF(0) values>90% and %Sign values>90% are highlighted in bold Table 13 (continued) OLS MM Variables Mean βMin βMax βØ se %Sign CDF(0) Mean βMin βMax βØ se %Sign CDF(0) L.dl.gerd − 0.03 − 0.39 0.51 0.23 0.3 58.9 − 0.01 − 0.08 0.20 0.06 0.4 78.8 L.dl.rend − 0.02 − 0.30 0.11 0.09 0.1 56.1 0.03 − 0.10 0.09 0.04 40.3 88.1 L.dl.hpi 0.04 − 1.18 1.82 0.52 6.8 55.2 − 0.11 − 0.27 0.09 0.15 0.0 91.8 L.dl.sugar − 0.06 − 0.42 0.74 0.14 23.8 53.2 0.02 − 0.05 0.06 0.03 0.6 93.2 190 K.Celebi et al. Acknowledgement We would like to thank participants of the 16th annual conference of the German Society of Health Economics (dggö) in Halle/Saale and an anonymous reviewerfor this journalfor valuable comments on an earlier draft of this paper. Sandqvist was affiliated at KOF Swiss Economic Institute, ETH Zurich, Switzerland and at ifo Institute—Leibniz Institute for Economic Research at the University of Munich, Germany when research for this paper was carried out. Funding Open Access funding enabled and organized by Projekt DEAL. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Declarations Conflict of interests None. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. References Angrist, J. D., & Pischke, J. (2010). The credibility revolution in empirical economics: how better research design is taking the con out of econometrics. Journal of Economic Perspectives, 24(2), 3–30. Barnett, V., & Lewis, T. (1994). Outliers in Statistical Data. Wiley. Bates, L. J., & Santerre, R. E. (2013). Does the U.S. health care sector suffer from Baumol’s cost disease? Evidence from the 50 states. Journal of Health Economics, 32(2), 386–391. Baumol, W. J. (1967). Macroeconomics of unbalanced growth: The anatomy of urban crisis. American Economic Review, 57(3), 415–426. Baumol, W. J. (2012). The cost disease. Why computers get cheaper and health care doesn’t. Yale University Press, New Haven (CT) and London. Bellido, H., Olmos, L., & Román-Aso, J. A. (2019). Do political factors influence public health expenditures? Evidence preand post-great recession. European Journal of Health Economics, 20(3), 455–474. Berndt, E. R., Cutler, D. M., Frank, R. G., Griliches, Z., Newhouse, J. P., & Triplett, J. E. (2000). Medical care prices and output. In A. J. Culyer, & J. P. Newhouse (Eds.), Handbook of health economics, vol. 1A. Amsterdam, Elsevier. Breusch, T. S. (1990). Modelling economic series. Oxford University Press. Castro, V. (2017). Pure, white and deadly … expensive: A bitter sweetness in health care expenditure. Health Economics, 26(12), 1644–1666. Colombier, C. (2012). Drivers of health care expenditures: does Baumol’s cost disease loom Large? FiFo Discussion Paper, No. 12–5. Colombier, C. (2017). Drivers of health-care expenditure: What role does Baumol’s cost disease play? Social Science Quarterly, 95(5), 1603–1621. Croux, C., Dhaene, G., & Hoorelbeke, D. (2008). Robust standard errors for robust estimators, mimeo. www. econ. kuleu ven. be. chris tophe. croux/ public. de la Maisonneuve, C., & Oliveira Martins, J. (2014). The future of health and long-term care spending. OECD Journal: Economic Studies, 61–96. Hartwig, J. (2008). What drives health care expenditure? – Baumol’s model of ‘unbalanced growth’ revisited. Journal of Health Economics, 27(3), 603–623. Hartwig, J., & Sturm, J.-E. (2014). Robust determinants of health care expenditure growth. Applied Economics, 46(36), 4455–4474. Hauck, K., & Zhang, X. (2016). Heterogeneity in the effect of common shocks on healthcare expenditure growth. Health Economics, 25(9), 1090–1103. 191 Baumol’s cost disease inacute versuslong‑term care: Do the… Helland, E., & Tabarrok, A. (2019). Why are the prices so damn high? Health, education, and the Baumol effect. Mercatus Center, George Mason University, Arlington, VA. Hlavac, M. (2016). ExtremeBounds: Extreme bounds analysis in R. Journal of Statistical Software, 72(9), 1–22. Jeetoo, J. (2020). Healthcare expenditure and Baumol cost disease in Sub-Sahara Africa. Economics Bulletin, 40(4), 2704–2716. Kirchgässner, G. (2009). Das schweizerische Gesundheitswesen: Kostenentwicklung und grundsätzliche Probleme. Die Volkswirtschaft, 11–2009, 4–8. Koller, M., & Stahel, W. A. (2011). Sharpening Wald-type inference in robust regression for small samples. Computational Statistics & Data Analysis, 55(8), 2504–2515. Kommission für Konjunkturfragen (2006). Reform des Gesundheitswesens, Jahresbericht, Bern, 8. September 2006. 385. Mitteilung, Beilage zum Magazin Die Volkswirtschaft (www. seco. admin. ch). Leamer, E. E. (1985). Sensitivity analyses would help. American Economic Review, 75(3), 308–313. Levine, R., & Renelt, D. (1992). A sensitivity analysis of cross-country growth regressions. American Economic Review, 82(4), 942–963. Lorenzoni, L., Marino, A., Morgan, D., & James, C. (2019). Health spending projections to 2030: new results based on a revised OECD methodology. OECD Health Working Papers No. 110, OECD Publishing, Paris. McAleer, M., Pagan, A. R., & Volker, P. A. (1985). What will take the con out of econometrics? American Economic Review, 75(3), 293–307. Medeiros, J., & Schwierz, C. (2013). Estimating the drivers and projecting long-term public health expenditure in the European Union: Baumol’s ‘cost disease’ revisited. European Commission DG EcFin Economic Papers No. 507. Newhouse, J. P. (1977). Medical-care expenditure: A cross-national survey. The Journal of Human Resources, 12(1), 115–125. OECD. (2008). OECD Glossary of statistical terms. OECD Publishing. OECD. (2023). Health at a glance: OECD indicators. OECD Publishing. OECD, Eurostat and World Health Organization. (2017). A system of health accounts 2011 (Revised). OECD Publishing. Roberts, J. (1999). Sensitivity of elasticity estimates for OECD health care spending: Analysis of a dynamic heterogeneous data field. Health Economics, 8(5), 459–472. Rossen, B., & Faroque, A. (2016). Diagnosing the causes of rising health-care expenditure in Canada: Does Baumol’s cost disease loom large? American Journal of Health Economics, 2(2), 184–212. Rousseeuw, P., & Yohai, V. (1984). Robust regression by means of S-Estimators. In: J. Franke, W. Härdle, & D. Martin (Eds.), Robust and nonlinear time series analysis, Lecture Notes in Statistics, vol 26, Springer, New York, NY. Sala-i-Martin, X. (1997). I just ran two million regressions. American Economic Review, 87(2), 178–183. Sala-i-Martin, X., Doppelhofer, G., & Miller, R. I. (2004). Determinants of long-term growth – a Bayesian averaging of classical estimates (BACE) approach. American Economic Review, 94(4), 813–835. Salibian-Barrera, M., & Yohai, V. (2006). A fast algorithm for S-regression estimates. Journal of Computational and Graphical Statistics, 15(2), 414–427. Swartz, S., & Welsch, R.E. (1986). Applications of bounded-influence and diagnostic methods in energy modeling. In: D. A. Belsley, & E. Kuh (Eds.), Model Reliability, MIT Press, Cambridge, MA. Temple, J. (2000). Growth regressions and what textbooks don’t tell you. Bulletin of Economic Research, 52(3), 181–205. Tian, F., Gao, J., & Yang, K. (2018). A quantile regression approach to panel data analysis of health-care expenditure in Organisation for Economic Co-operation and Development countries. Health Economics, 27(12), 1921–1944. Vandermeulen, A. (1968). A remission from Baumol’s disease: Ways to publish more articles. Southern Economic Journal, 35(2), 189–191. Verardi, V., & Croux, C. (2008) Robust regression in Stata, KU Leuven, Working Paper KBI 0823. Wang, L., & Chen, Y. (2021). Determinants of China’s health expenditure growth: based on Baumol’s cost disease theory. International Journal for Equity in Health, 20(1), Article 213. Yohai, V. (1987). High breakdown-point and high efficiency estimates for regression. The Annals of Statistics, 15(2), 642–665. Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.