Aims Consumer Valuation of Medicare Part D Plans • How much social welfare has been created by this new government program? – consumer surplus, producer surplus, and net social surplus • What features of the drug plans were valued by consumers, and by how much? max uijt = α ( yi − p jt ) + X 'jt β + ξ jt + ε ijt j∈{0,..., J t } Claudio Lucarelli Dept of Policy Analysis and Management, Cornell University Jeff Prince Dept of Applied Economics and Management, Cornell University Kosali Simon (presenter) Dept of Policy Analysis and Management, Cornell University and NBER Motivation • Plan features are heavily directed by policy, thus relevant to know how they are valued by consumers. • Important to know whether the cost of the program is less than the surplus created, and the split between industry and consumers, to know whether the program is a “give-away” to the industry. Method-Discrete Choice Analysis Prior Work Using Similar Method • Demand estimation in health care markets – Town and Liu (RAND, 2003) find substantial net social welfare from M+C, particularly from the drug benefits. • Follows Berry (1994) • By studying aggregate market shares, we recover parameters of the underlying utility function Allows us to estimate the value derived from plan characteristics, the price (and other features) elasticity of demand (including cross price elasticities of demand), CS, PS • • Forecasts that Medicare Part D will create similar surplus • Data and Method • Data: CMS landscape files, plan enrollment file web queries – Enrollment reported by CMS includes those who did not chose a plan but were auto-enrolled – We calculate the number of enrollees who voluntarily selected into a plan – In addition to plan characteristics from the CMS landscape file, we calculate OOP for set of drugs N=1,429 • Method: Final model Scaled market share= log(mktshare) log(outsidemktshare)= f(premium,gap, Oop, deductible, top100 on formulary, PA, cs under $20), with insurer fe, region fe – utility function U=g(x,e) – Behavioral process that leads to a choice is y=f(x,ε) – ε is unobserved [1] P(y | x) = P(ε s.t. f(x, ε) = y) If errors are iid type 1 extreme value, [1] has a logistic distribution; we estimate a logit for transformed market shares, y), includes insurer f.e., region f.e. This implies all plans independent of each other. Means that all cross price elasticities depend only on prices and market shares of the two plans In reality, demand may be nested (i.e. people decide whether to go into basic or enhanced plans), implies a nested logit Supply side: We assume Bertrand pricing by firms, Descriptive Statistics Variable Mean St.Dev Unadjusted enrollment 10,932 25,389 Adjusted enrollment 7,010 20,687 Scaled market share 0.025 0.065 # top 100 drugs on formulary 93 6.62 Monthly premium $37.4 12.85 Annual OoP top 5 $753.08 339.53 Annual deductible $92.24 115.79 1 Estimates from Simple Logit Variable Coefficient Interpretation of coefficients: Gap coverage 0.50 *** Out of pocket index -0.0007 ** Premium -0.07*** Deductible -0.009*** Top 100 drugs on formulary 0.095 *** Prior auth top 100 0.008 Top 100 under $20 0.01226 *** -.07*y*(1-y)=% change in scaled market share as my own price increases 1% -.07*y1*y2=cross price elasticity .5/.07=$7.14 is value placed on gap coverage pm Other specifications • Without insurer fixed effects • With different OOP measures • With nest (gap coverage or not) Includes fe for region and insurer, * Indicates statistical significance at te 10% level, ** at the 5% level and *** at the 1% level Selected Estimates from Nested Logit vs Simple Logit Simple logit Nested logit Variable Coefficient Coefficient Gap coverage 0.50 *** -- OOP1 -0.0007 ** -0.0007** Premium -0.07*** -0.085*** Deductible -0.009*** -0.0097 *** Top 100 drugs on formulary 0.095 *** 0.099** Prior auth top 100 0.008 -0.001 Top 100 under $20 0.01226 *** 0.012*** Very preliminary: Consumer welfare with and without enhanced plans With: $8,954,700 • Without: $5,786,100 • % Difference: Approx. 35% lower without gap coverage Includes fe for region and insurer, * Indicates statistical significance at te 10% level, ** at the 5% level and *** at the 1% level Conclusions • The value of each characteristic is shown by coefficients from logit estimation – When own premium increases by 1%, own demand (enrollment) decreases by roughly 34% • Importance of demand estimation vs OLS – OLS coef on regression ms=f(X) is interpreted as difference in enrollment from x increasing but cannot tell us about structural parameters of utility 2