Evaluation of Parity in the Federal Employees Health Benefits (FEHB) Program: Final Report. Appendix A: Detailed Model Specification for Plan Exit and Carve-out Analyses

12/31/2004

Plan Exit Analysis

The Generalized Estimating Equations (GEE) model used to estimate the probability of plan exit and the change in this probability from 2000 to 2001 and 2002 was:

logit (pr (EXIT))pt = α0 + α1YEAR01t + α2YEAR02t + α3PLANTYPEp + α4VLIMITS(LESS)p + α5VLIMITS(MORE)p + α6DLIMITS(LESS)p + α7DLIMITS(MORE)p + α8CSHAREp + α8MEDCSHAREp + α9ENROLLp + α10REGION1p + α11REGION2p + α12REGION3p + α13REGION4p + α14 YEAR01t*VLIMITS(LESS)p + α15YEAR01t*VLIMITS(MORE)p + α16YEAR02t*VLIMITS(LESS)p + α17YEAR02t*VLIMITS(MORE)p + α18YEAR01t*DLIMITS(LESS)p + α19YEAR01*DLIMITS(MORE)p + α20YEAR02t*DLIMITS(LESS)p + α21YEAR02t*DLIMITS(MORE)p    (1.1)
in which

  • EXIT indicates whether a health plan exited the FEHB Program,
  • p indexes health plans,
  • t indexes time,
  • YEAR01 and YEAR02 are year dummies indicating whether the time period was either post-parity year (2001 or 2002) compared with 2000, the pre-parity year,
  • PLANTYPE indicate whether a plan was an HMO or FFS,
  • VLIMITS(LESS) indicates whether a health plan had less restrictive outpatient limits of between 31 and 60 visits a year compared to no limits before parity,
  • VLIMITS (MORE) indicates whether a health plan had more restrictive outpatient limits of between 20 and 30 visits a year compared with no limits,
  • DLIMITS(LESS) indicates whether a health plan had less restrictive inpatient day limits of between 31 and 60 days a year compared to no limits before parity,
  • DLIMITS(MORE) indicates whether a health plan had more restrictive inpatient day limits of between 20 and 30 days a year compared with no limits,
  • CSHARE indicates a copayment equal to or greater than $25 and coinsurance equal to or greater than 40 percent for outpatient mental health services in 2000,
  • MEDCSHARE indicates a general medical copayment of either greater than or less than $10,1
  • ENROLL refers to FEHB Program plan enrollment size measured at a single point in time (2002)2, and
  • four REGION dummies indicate whether a health plan was available nationwide or located in one of three regions (Northwest, Midwest, and South) compared with an omitted region, the West.

Carve-out Analyses

The GEE model used to estimate the probability of plan carve-out and the change in this probability from pre- to post-parity, as well as the comparison of this probability for FEHB versus Medstat plans, was:

logit (pr (CO))pt= α0 + α1POSTt + α2PARITYp + α3PLANTYPEp + α4ENROLLp + α5REGION1p + α6REGION2p + α7REGION3p + α8REGION4p + α9POSTt*PARITYp    (1.2)

  • in which
  • CO, the outcome variable, indicates the probability of carving out,
  • p indexes plans,
  • t indexes time,
  • POST indicates whether the time period was before or after parity implementation,
  • PARITY indicates whether a plan was an FEHB plan or a Medstat plan,
  • PLANTYPE indicates whether the plan was a FFS/PPO (preferred provider organization) or an HMO/POS (point of service) plan, and
  • the enrollment and region variables are defined as in Equation 1.1.

Since this estimation model was non-linear, the PERT could not interpret the coefficient of the interaction of POST and PARITY in Equation 1.2 directly, but rather computed the cross difference as indicated in 1.3 and the standard errors using the approach developed by Ai and Norton (2003). Predicted probabilities
of carving out were calculated as:

Diff − in − Diff = (Ŷ1,1 − Ŷ1,0) − (Ŷ0,1 − Ŷ0,0)    (1.3)

in which Ŷ1,1 was defined as the predicted probability of carving out among FEHB plans after parity in 2001 (parity equals 1, post equals 1). In comparison, Ŷ1,0 indicated the predicted probability of carving out among FEHB plans before parity in 2000 (parity equals 1, post equals 0). Similarly, Ŷ0,1 was defined as the predicted probability of carving out among Medstat comparison plans in 2001 (parity equals 0, post equals 1), and Ŷ0,0 was defined as the predicted probability of carving out among Medstat plans in 2000 (parity equals 0, post equals 0). This calculation produced a net impact estimate of the change from before to after parity in the predicted probability of carving out among FEHB plans as a percentage of the change among Medstat plans.
The GEE model used to estimate the probability of plan carve-out and the change in this probability from pre- to post-parity was:

logit (pr (CO))pt = α0 + α1POSTt + α2ASSOCIATIONp + α3VLIMITS(LESS)p + α4VLIMITS(MORE)p + α5DLIMITS(LESS)p + α6DLIMITS(MORE)p + α7CSHAREp + α8ENROLLp + α9REGION1p + α10REGION2p + α11REGION3p + α12REGION4p + α13POSTt*ASSOCIATIONp + α14POSTt *VLIMITS(LESS)p + α15POSTt*VLIMITS(MORE)p   (1.4)
in which

  • CO indicates the probability of carving out,
  • p indexes health plans,
  • t indexes time,
  • POST indicates whether the year was before or after parity implementation,
  • ASSOCIATION indicates whether the plan was an Association or non-Association plan, pre-parity visit limit, day limit, and cost sharing dummies in 2000 were constructed as in Equation 1.1, and
  • the enrollment and region variables are constructed as in Equation 1.1.

For this model, a variable for plan type was not included because of the high correlation between plan type and the Association dummy. In addition, in Equation 1.4, the pre-parity visit limit, day limit, and cost sharing dummies in 2000 captured benefit information in 2000 to determine whether pre-parity benefits affected carving out.3


1 Results were qualitatively similar testing different cut-points for construction of limit and cost-sharing variables.

2 Relative plan enrollment did not tend to vary substantially across years.

3 Separate analyses (results not presented) using 1999 health plan pre-parity benefit variables rather than 2000 pre-parity benefit variables produced similar results.

View full report

Preview
Download

"parity.pdf" (pdf, 5.07Mb)

Note: Documents in PDF format require the Adobe Acrobat Reader®. If you experience problems with PDF documents, please download the latest version of the Reader®

View full report

Preview
Download

"parityA.pdf" (pdf, 61.31Kb)

Note: Documents in PDF format require the Adobe Acrobat Reader®. If you experience problems with PDF documents, please download the latest version of the Reader®

View full report

Preview
Download

"parityB.pdf" (pdf, 194.85Kb)

Note: Documents in PDF format require the Adobe Acrobat Reader®. If you experience problems with PDF documents, please download the latest version of the Reader®

View full report

Preview
Download

"parityC.pdf" (pdf, 134.89Kb)

Note: Documents in PDF format require the Adobe Acrobat Reader®. If you experience problems with PDF documents, please download the latest version of the Reader®

View full report

Preview
Download

"parityD.pdf" (pdf, 147.52Kb)

Note: Documents in PDF format require the Adobe Acrobat Reader®. If you experience problems with PDF documents, please download the latest version of the Reader®