Sa4E9 - International Health Comparisons

Front 1

Statistics: Outline

Back 1

1. Introduction
2. Cross-sectional analysis
3. Panel Regression Analysis
4. Stationarity & Cointegration
5. Conclusion

Front 2

Introduction

Back 2

1. Trends in HC expenditure
2. Research Question

Front 3

Intro: Trends in HCE

Back 3

- proportion of health care expenditures to aggregate income has dramatically increased over time
- large variation in absolute and relative health expenditures
- diminishing returns

Front 4

Intro: Research Question

Back 4

- What determines the quantity of resources a country devotes to medical care? (sig, size, direction)
- institutional: HC org, finan
- non-inst: dem, tech
- allocatively efficient levels of health care expenditures?
- Positive question with normative implications

Front 5

Cross-sectional Analysis

Back 5

a. Newhouse (1977)
b. Parkin (1987)
- weak theoretical basis
- Methodology
- role of price
- functional form
c. Cross-sectional v. Panel

Front 6

Cross-sectional: Newhouse (1977)

Back 6

- model specification:
- What determines the quantity of resources a country devotes to medical care?
- Aggregate income is the primary determinant of health expenditure (over 90 per cent of variation as measured by the coefficient of determination (R2))
- Implies the production of health from health care & organization of heatlh system of only secondary concern

Front 7

Cross-sectional: Parkin et al (1987) - weak theoretical basis

Back 7

o implicit formulation of macreeconomic demand as a function of microeconomic demand by interpreting aggregated data through the microeconomic theory of the Engel curve despite limited development of economic theory linking the two demand functions
o Engel curve: relationship between a consumer’s income and the quantity of the good purchased, holding tastes and preferences constant
o Income elasticity – ratio of marginal propensity to consume (the slope of the Engel curve) to the average propensity to consume
o Assuming fixed prices, the income elasticity can be estimated in terms of expenditure rather than quantity demanded (price x quantity = expenditure)

Front 8

Cross-sectional: Parkin et al (1987) - methodology - role of prices

Back 8

• Newhouse (1977): relative health care factor prices do not vary with aggregate income.
• McGuire (2008): labor costs comprise 70% of health expenditure and that labor costs are higher in high income countries.
• Newhouse (1977): exchange rates
• Parkin, et al (1987): PPP → 1.12 to 0.90
• Challenges of PPP for non-market commodities eg health care

Front 9

Cross-sectional: Parkin et al (1987) - methodology - functional form

Back 9

• Linear function form
• Elasticity > 1 for all positive income levels with a negative intercept therefore significance of intercept is also important
(Newhouse’s α = -60, -51, does not report the p-value)
(Parkin et al do not have sig alpha either)
• Error – omitted variables (correlated with aggregate income) introduce bias into the coefficient estimation with the sign of the correlation
• Positive correlation between the relative factor prices and income would introduce a positive omitted variable bias True income elasticity is lower than estimated
• Univariate cross-sectional regression analysis underspecified introducing significant omitted variable bias because does not control for institutional and non-institutional determinants of health care expenditure
• Constant unity elasticity functional form – best empiric fit

Front 10

Cross-sectional: cross sectional v. panel - criticisms

Back 10

o Summary of criticisms:
o over-aggregated data
o misspecified functional forms
o atheoretical nature of analysis
o Statistical criticisms:
o assumes homogeneous relationships between determinants and health expenditures despite heterogeneous preferences and production functions
o assumes static relationships between determinants and health expenditures
o trade-offs between study power and data quality

Front 11

Cross-sectional - cross-sectional v. panel - panel data

Back 11

o Incorporates country- and time-effects: heterogenous & dynamic relationships
o Increases sample size by adding time-series data – increase statistical power

Front 12

Panel regression analysis

Back 12

vector autoregressive models
- decomposition of error term
- assumptions re: error term --> model specifications
Empirical evidence
1. Gerdtham (1992)
2. Gerdtham et al., (1998)
3. Barros (1998)

Front 13

Panel regression analysis - vector autoregressive models - error term decomposition

Back 13

o stochastic error term, εit
o country specific error term, μi
o time-specific error term, θt

Front 14

Panel regression analysis - vector autoregressive models - assumptions re: error

Back 14

o E(μI, θt) = 0 → OLS coefficient estimation
o E(μI, θt) = constant: time-invariant process across units and are being captured by the model but can be modeled through the error term alone → fixed effects model (reject prior model via F-test of the joint significance of country- and time-specific dummy variables) – first difference model
Takes the fixed effects for stationarity
o μi and/or θt: random variables – random error components model – GLS; f estimation of coefficients via Lagrange multiplier test or Hausman test appropriateness of fixed effects or random-effects models

Front 15

Empirical evidence: Gerdtham 1992

Back 15

o Gerdtham (1992) – fixed effects and random effects model in panel data
• Time- and country- specific fixed effects model best fit data
• Estimated income elasticity of 0.74
• None of explanatory variables were statistically significant - time and country specific dummy variables may have weakened interpretation

Front 16

Panel regression analysis - Empirical evidence on institutional variables

Back 16

ambiguous: share of public financing, gatekeeping, public contracts v. integrated
budget ceiling: + Gerdtham et al
BUT system variables may be endogeneous

Front 17

Panel regression analysis - Empirical evidence - Barros 1998

Back 17

o Barros (1998) – y = health care expenditure growth
• Sig: initial health expenditure (+), initial health expenditure squared (-), growth of GDP
• Negative effect of initial expenditure on health expenditure growth is highest in countries with high initial health expenditure
• Income elasticity – slightly below one

Front 18

Stationarity & Cointegration

Back 18

1. stationarity
a. non-stationarity - defn
b. non-stationarity - problems
c. Non-stationarity + cointegrated variables
2. Empiricial evidence
a. Non-stationary
b. Stationary

Front 19

Stationarity & Cointegration - non-stationarity - defn

Back 19

- assumption of stationarity – probability distribution of the variable does not change over time
- if stationary → see above – although the fixed effects model takes the first difference
- non-stationarity – persistent long-term movement of a variable over time
o deterministic trend – non-random function of time
o stochastic trend – random and varies over time
o Augmented Dickey Fuller Test (ADF) → unit root → stochastic trend →
o Keep differencing – harder to interpret the coefficients

Front 20

Stationarity & Cointegration - 3. non-stationarity - problems

Back 20

- OLS estimation of non-stationary variables with stochastic trends →
o Biased coefficient estimates → bias in income elasticity estimate
o Non-normal distribution of t-statistic → altering hypothesis testing and confidence interval calculation → spurious regression

Front 21

Stationarity & Cointegration -
4. Non-stationarity + cointegrated variables

Back 21

- Non-stationary + cointegrated variables –
o Variables share a common trend over time
o Difference in non-stationary trends are stationary
o Vector error correction model

Front 22

Stationarity & Cointegration -
5. Empiricial evidence

Back 22

• ADF: Non-stationary (McCoskey & Selden (1998) – stationary – conclusions compromised by omission of time trend)
o Cointegrated? Using varied stat tests for cointeration:
• Hansen & King (1996): x cointegrated in static model → spurious relationaship
• Blomqvist & Carter (1997): cointegrated in dynamic model
• Roberts (1998b): x cointegrated in static model

Front 23

Statistical relationships conclusions

Back 23

- economic v. normative; substantive interpretation?
- Specify relationship between relative price of health care and quantity demanded as it may be endogenous
- Errors in measurement bias – poor quality data sets
- Small statistically sig effects of gatekeepers, patient reimbursement by insurer, capitation reimbursement of ambulatory care providers, ratio of inpatient to outpatient care, total supply of physicians, budget ceilings, and public sector provision of health services on health care expenditures
- new explanatory variables: gov budget deficit, tax subsidy of private health insurance
- better tests for non-stationarity and cointegration

Front 24

Policy implications of Luxury good:

Back 24

1. Newhouse (1977)
2. Parkin et al (1987)

Front 25

Policy implications of Luxury good: Newhouse (1977)

Back 25

- estimated income elasticity > 1 → luxury good
- “the marginal unit of medical care” (1977, p.120) “may do little or nothing for mortality and morbidity rates, it may well produce improvements in so-called subjective components of health,” (1977, p.122).
- institutions = endogenous variables – wealthier nations may implement institutions more efficient at improving subjective health
- Microeconomic analysis - household level data: income elasticity estimates < 1
- Newhouse: distortion of health care price at point of consumption; nation faces the full cost of health care expenditure and may rely on non-price rationing mechanisms, such as waiting time, to influence individual demand.

Front 26

Policy implications of luxury good: Parkin et al (1987)

Back 26

- Interpretation: distinguish between the microeconomic description of patterns of behaviour in terms of luxury and necessary goods from normative ‘luxury’ of medically essential and medically frivolous health care consumption.
- Demand-side policies re: cost-sharing return a + marginal cost to consumption (co-insurance, co-payment, deductible) → ↓ demand of effective & ineffective care; cost-shifting to non-consumer initiated sectors (eg inpatient care)
- If price elasticity of demand < 1 → cost-shifting economic normal good → highlight need for financial protection of lower income groups as medical care expenditure would be a disproportionate proportion of their budget

Front 27

Technology and HCE growth - Short

Back 27

1. Weisbrod
2. Newhouse
3. Barros

Front 28

Technology and HCE growth - Long 1

Back 28

Weisbred – two way causal relationship HC tech and HI
- HC ins – retrospective → prospective: average treatment costs
- Cost-increasing technology -> incr avg cost, incr variance of cost -> motivates broaden HI coverge

Front 29

Technology and HCE growth - Long 2

Back 29

Newhouse
- demand influences: demographic aging, insurance demand, income growth
- supply influences: SID, productivity differentials
- y = growth in HCE
- residual approach
- over 5 years in US, 50% of HCE growth attributable to new tech
- over 10 years in Australia, 17-55% of HCE growth attributable to HC

Front 30

Technology and HCE growth - Long 3

Back 30

Barros
- covergence: dispersion in the variations in the level of HCE across countries fell during the 1960-1970s and has been constant since
- Med tech diffuses among countries
- Health care technology explains 30% of the growth in OECD health expenditure growth
- Countries with lower levels of HCE have faster growth
- Catch up over time
- The negative of effect of initial health expenditure on growth of HCE is greater that greater the initial income
- Regression results