Memo Template - Download Now DOC by 5LJ1hN1

VIEWS: 14 PAGES: 9

									MEMORANDUM                                                                        MATHEMATICA
                                                                                  Policy Research, Inc.
                                                                                  600 Maryland Ave. S.W., Suite 550
                                                                                  Washington, DC 20024-2512
                                                                                  Telephone (202) 484-9220
                                                                                  Fax (202) 863-1763
                                                                                  www.mathematica-mpr.com

TO:          Allen Berkowitz and Stephen Meskin


FROM:        Jonathan Jacobson                                                      DATE: 87/125/2003
                                                                                    8986-0053

SUBJECT:     Conceptual Framework for Simulating Health Care Priority
             Status by State for VAM3 (REVISED)


        In this memo, I outline a possible conceptual framework for simulating veteran health
        care priority status (HCP) by state for the 2003 Veterans Actuarial Model (VAM3).
        This framework is based on the VAM retreat held on July 2, 2003, at the office of Hay
        Management Consultants in Arlington, Virginia, and on subsequent comments from the
        VA Office of the Actuary.

        I first review the design requirements of the proposed model database. Next, I discuss
        the datasets that would need to be created before creating this database. I then present a
        sequence of steps by which we could construct the database with its underlying
        classification variables, and determine the parameters for assigning veterans to states. I
        then describe the other parameter files that would be impacted by the proposed
        redefinition of the model’s classification variables. I conclude with a discussion of next
        steps in model development, assuming this framework is endorsed by VA.


        1. Design Requirements of the Model Database

        The proposed model database will include both 4/1/2000 and 9/30/2002 baselines. The
        4/1/2000 baseline will consist of (i) data on veterans who were age 49 or older as of
        4/1/2000 and who did not serve after the Vietnam War, obtained primarily from Census
        2000, and (ii) data on veterans under age 49 as of 4/1/2000 or who served after the
        Vietnam War, obtained primarily from DMDC sources. The model database will
        include the following classification variables:

             SOURCE = source of data (Census or DMDC)

             DATE = date of projection (4/1/2000 – 9/30/2102)

             AGE = age of veteran (17-120)

             GENDER = gender of veteran (Male or Female)
MEMO TO:   Allen Berkowitz and Stephen Meskin
FROM:      Jonathan Jacobson
DATE:      87/125/2003
PAGE:      2



        PERSVC = period of service (VAM2 codes to be used)

        LIVING = living/deceased status

        RACE = race/ethnicity (exact codes to be determined, since Census allows
         multiple responses while older DMDC data did not)

        HCP_STAT = health care priority status

   HCP_STAT will consist of nineten possible values, hierarchically defined such that
   each lower category will exclude veterans in higher priority groups. Values of the
   variable will correspond with information available from either the relevant
   Compensation and Pension (C & P) mini master frozen files (3/31/2000 will be used for
   the 4/1/2000 baseline, with 3/31/2001, 9/30/2001 and 9/30/2002 files also available), or
   from Census 2000 data

      1. 1AD = veterans with 70% or higher degree of service-connected (SC) disability
         as indicated in the C & P data (corresponds with Health Care Enrollment
         Priority 1A)

      2. 1EF = veterans with 50 to 60 % degree of SC disability as indicated in the C &
         P data (corresponds with Health Care Enrollment Priority 1B)

      3. 2GH = veterans with 30 to 40 % degree of SC disability as indicated in the C &
         P data (corresponds with Health Care Enrollment Priority 2)

      4. 3IJ = veterans with 10 to 20 % degree of SC disability and a service-connected
         disability as indicated in the C & P data (corresponds with a portion of the
         veterans in Health Care Enrollment Priority 3)

      5. 3K_4L_5M = veterans with pension income as indicated in the C & P data
         (corresponds with a portion of veterans in Health Care Enrollment Priority 3, a
         portion of veterans in Health Care Enrollment Priority 4, and a portion of
         veterans in Health Care Enrollment Priority 5)

      6. 5N = low-income veterans as indicated in the Census data (corresponds with a
         portion of veterans in Health Care Enrollment Priority 5)

      7. 6O = veterans with compensable 0 % degree of SC disability as indicated in the
         C & P data (corresponds with a portion of the veterans in Health Care
         Enrollment Priority 6)
MEMO TO:   Allen Berkowitz and Stephen Meskin
FROM:      Jonathan Jacobson
DATE:      87/125/2003
PAGE:      3



      8.6P = veterans who served in or after the Gulf War as indicated in the Census and          Formatted: Bullets and Numbering
         DMDC data (corresponds with a portion of veterans in Health Care Enrollment
         Priority 6)

      9.8.6P_7Q = veterans with incomes falling at or below the HUD geographic index              Formatted: Bullets and Numbering
          as indicated in the Census data (corresponds with a portion of veterans in Health
          Care Enrollment Priority 6 and all veterans in Health Care Enrollment Priority
          7)

      10.9. 6P_8R = veterans with incomes above the HUD geographic index as                       Formatted: Bullets and Numbering
         indicated in the Census data (corresponds with a portion of veterans in Health
         Care Enrollment Priority 6 and all veterans in Health Care Enrollment Priority
         8)


   We believe that, by not distinguishing degree of disability any further in the
   classification variables than is outlined above, we will prevent the model database from
   becoming so large as to be unmanageable. For the sake of simulating benefit utilization
   and precise health care priority enrollment status, parameters (a revised version of
   NEWDEGR) will be used to split HCP_STAT = 1AD by degree of disability (70, 80,
   90, or 100), HCP_STAT = 1EF by degree of disability (50 or 60), HCP_STAT = 2GH
   by degree of disability (30 or 40), HCP_STAT = 3IJ by degree of disability (10 or 20),
   and HCP_STAT = 3K_4L_5M by health care enrollment priority (priority groups 3, 4,
   or 5). Additional parameters in HCP_6P will use period of service information to
   identify veterans with HCP_STAT = 7Q or 8R who should be assigned to health care
   priority group 6 in order to represent veterans exposed to toxic substances or radiation.

   The revised NEWDEGRT parameters will enable users to determine what proportion of
   veterans with HCP_STAT = 3K_4L_5M are in health care priority group 3
   (corresponding with POWs, Purple Heart recipients, those discharged from duty
   because of a disability or who became disabled because of treatment or vocational
   rehabilitation), what proportion are in health care priority group 4 (receiving aid and
   attendance or housebound benefits, or catastrophically disabled), and what proportion
   are in group 5 (other veterans receiving VA pension benefits). In addition, new
   HCP_6P parameters will determine what proportion of veterans with HCP_STAT =
   6P_7Q or 6P_8R should in fact be assigned to health care priority group 6. We believe
   that using parameters rather than classification variable values to assign veterans to
   these specific priority groups will give VA flexibility to test different assumptions rather
   than “hard-code” questionable assumptions in the database itself when the relevant
   cross-distributions (that is, distributions with the other classification variables) are
   unknown.
MEMO TO:   Allen Berkowitz and Stephen Meskin
FROM:      Jonathan Jacobson
DATE:      87/125/2003
PAGE:      4



   Not included in this list of classification variables is Length of Service (LOS), which
   could continue to function as a classification variable for national simulations, even if
   memory requirements made it difficult to continue to form this classification variable
   for state simulations.

   State will be determined by the parameter file SPVPD, as it is presently, except that, in
   addition to varying by GENDER and AGE, state will also vary by PERSVC, RACE,
   and HCP_STAT. This variation will allow tabulations of veterans by state to be
   accurate when they are done by any of these classification variables.


   2. Precursor Datasets to the Model Database

   As prerequisites to creating the 4/1/2000 model database, it will be necessary to create
   the following datasets: (i) an extract from Census 2000 data, (ii) an extract from
   DMDC data, including information on veterans from both the Active Duty Loss File
   and the Reserve File; and (iii) an extract based on the C & P mini master frozen files.
   The key variables included in each of these files are listed in Table 1.

   The Census 2000 dataset will need to contain the number of veterans for unique cells
   defined by each of six variables: GENDER, AGE, PERSVC, RACE, STATE, and
   income category (INC_CAT). Income category will have three four possible values: 0
   = below the pension threshold; 1 = not below the pension threshold but low income; 2
   = not low income but with income at or below the corresponding HUD geographic
   index; and 3 = income at a level above the level of the HUD geographic index. Census
   includes some disability information but not information on degree of disability. It is
   possible that the counts in the Census dataset will need to be “raked” to correct for
   rounding errors and ensure that certain distributions meet control totals defined by
   published Census statistics on veterans.

   The DMDC dataset will contain, for veterans under age 49 as of 4/1/2000 or serving
   after the end of the Vietnam War, the number of veterans for unique cells defined by
   each of four variables: GENDER, AGE, PERSVC, and RACE. The main advantage of
   these counts is that they will permit adjustments for possible Census miscounts or the
   non-counting of veterans living overseas. (Note that INC_CAT and DIS_CAT are not
   contained in the DMDC data, and the STATE information is unreliable because it only
   indicates the state of residence at the time of separation.)

   The C & P dataset will contain, for veterans receiving compensation or pension
   benefits, the number of veterans for unique cells defined by each of five variables:
   GENDER, AGE, PERSVC, STATE, and disability category (DIS_CAT). PERSVC
MEMO TO:    Allen Berkowitz and Stephen Meskin
FROM:       Jonathan Jacobson
DATE:       87/125/2003
PAGE:       5


                                           TABLE 1

        VARIABLES TO INCLUDE IN PRECURSOR DATASETS TO 4/1/2000 MODEL
                                 DATABASE


Variable                                 Census 2000            DMDC                C&P
GENDER                                       X                   X                    X
AGE                                          X                   X                    X
PERSVC (Period of Service)                   X                   X                     1
                                                                                     X
RACE (Race/Ethnicity)                         X                    X                  --
STATE                                         X                    --                  1
                                                                                     X
INC_CAT (Income Category)                     X                    --                 --
DIS_CAT (Disability Category)                 --                   --             X [no “7”]


    and STATE are considered to be weakly estimated in the C&P1, so comparisons will be
    made with Census totals for data plausibility. DIS_CAT will have seven possible
    values: 1 = 70 to 100 percent degree of SC disability, 2 = 50 to 60 percent degree of
    SC disability, 3 = 30 to 40 percent degree of SC disability, 4 = 10 to 20 percent degree
    of SC disability, 5 = receipt of pension income without a service-connected disability
    and degree of disability above zero, 6 = compensable zero percent degree of SC
    disability and no pension income, and 7 = no disability or pension income.


    3. Steps for Creating the Model Database and State Distribution Parameters

    The first step in creating the 4/1/2000 database will be to adjust the counts of veterans
    in the Census 2000 dataset to match the counts in the DMDC data. Counts will be
    formed, in each dataset, of the number of veterans for the unique cells defined by
    GENDER, AGE, PERSVC, and RACE. (Note that these are the variables common
    between the Census and DMDC data.) For veterans under age 49 as of 4/1/2000 or
    who served after the Vietnam War, the number of veterans in each cell so defined will
    be scaled by the ratio of the number of veterans in the DMDC cell to the number of
    veterans in the Census cell. Cells in the Census for veterans age 49 or older as of
    4/1/2000 (and not serving after the Vietnam War) will be scaled by the ratio of the total

    1
      C&P data do not record multiple periods of service but record PERSVC as the first wartime
period for a pension recipient, and the period during which a SC disability occurred for a comp
recipient. The C&P measure of STATE is the state to which the check is sent, not necessarily
the state of residence.
MEMO TO:   Allen Berkowitz and Stephen Meskin
FROM:      Jonathan Jacobson
DATE:      87/125/2003
PAGE:      6


   number of DMDC veterans under age 49 (or who served after the Vietnam War) to the
   total number of Census veterans under age 49 (or who served after the Vietnam War),
   thereby adjusting for any possible Census miscount and for the exclusion of overseas
   veterans from the Census. (The relevant SAS program will keep track of the size of
   these adjustments to permit an additional explicit adjustment for the number of
   overseas veterans using the CENSUNDER parameters.) Because the Census data also
   include state and income category information missing from the DMDC data, the
   income category and state distribution within each group of cells defined by GENDER,
   AGE, PERSVC, and RACE will remain unchanged. At the end of the rescaling, the
   new dataset will be described as the merged Census-DMDC dataset.

   The second step in creating the 4/1/2000 database will be to compare the merged
   Census-DMDC data with the unmatched C & P data extract for the unique cells defined
   by GENDER, AGE, PERSVC, and STATE. Cells for which the C & P total exceeds
   the rescaled Census total will be flagged and then rescaled in the merged Census-
   DMDC data to match the C & P total. After this rescaling, the Census-DMDC veterans
   in each unique cell defined by GENDER, AGE, PERSVC, and STATE will be assigned
   to values of DIS_CAT in proportions that will replicate the corresponding counts
   contained in the C & P data. Because the Census-DMDC data will include
   race/ethnicity and income category information missing from the C & P data, the
   race/ethnicity and income category distribution within each group of cells defined by
   GENDER, AGE, PERSVC, and STATE will remain unchanged, and the distribution of
   DIS_CAT will be assumed to be the same for persons in each group of cells so defined,
   regardless of their race/ethnicity or income category.

   Before saving the new dataset as the 4/1/2000 baseline file, a final adjustment will be
   necessary to create the HCP_STAT variable from the DIS_CAT, PERSVC, and
   INC_CAT variables. As indicated in Table 2, HCP_STAT will be set equal to 1AD if
   DIS_CAT = 1, else 1EF if DIS_CAT = 2, else 2GH if DIS_CAT = 3, else 3IJ if
   DIS_CAT = 4, else 3K_4L_5M if DIS_CAT = 5, else 5N if INC_CAT <= 1, else 6O if
   DIS_CAT = 6, else 6P if PERSVC indicates a Gulf War or later period of service, else
   6P_7Q if INC_CAT = 2, else 6P_8R. After this adjustment, unique cells will be
   formed by GENDER, AGE, PERSVC, RACE, and HCP_STAT, and for each
   combination of cells, the distribution of veterans for each value of STATE will be
   determined. The state distributions will constitute the new parameters of the SVPD
   parameter file for allocating veterans to states as of 4/1/2000 according to their
   GENDER, AGE, PERSVC, RACE, and HCP_STAT. The file containing national
MEMO TO:    Allen Berkowitz and Stephen Meskin
FROM:       Jonathan Jacobson
DATE:       87/125/2003
PAGE:       7


                                         TABLE 2

     ASSIGNMENT OF HCP_STAT BASED ON DIS_CAT, PERSVC, AND INC_CAT


                                   INC_CAT< =          INC_CAT = 2          INC_CAT = 3
DIS_CAT and PERSVC                    1 (low            (<= HUD)              (>HUD)
                                     income)
DIS_CAT = 1, 70-100 % SC               1AD                 1AD                  1AD
DIS_CAT = 2, 50-60 % SC                1EF                 1EF                  1EF
DIS_CAT = 3, 30-40 % SC                2GH                 2GH                  2GH
DIS_CAT = 4, 10-20% SC                  3IJ                 3IJ                  3IJ
DIS_CAT = 5, Pension                3K_4L_5M            3K_4L_5M             3K_4L_5M
DIS_CAT = 6, Comp 0 % SC                5N                  6O                   6O
DIS_CAT = 7, other PERSVC               5N                6P_7Q                6P_8R


    counts of veterans by GENDER, AGE, PERSVC, RACE, and HCP_STAT will become
    the portion of the 4/1/2000 baseline database corresponding with living veterans; the
    portion corresponding with deceased veterans will be formed in the same manner as for
    the VetPop2001 model.

    HCP_STAT will not provide all of the detail model users will need to determine health
    care priority group or degree of disability. Finer distinctions within HCP_STAT
    categories will be determined by parameters (NEWDEGR AND HCP_6P, described
    below) instead of classification variables.


    4. Other Parameter Files Impacted by Redefining the Model Database

    Replacing DIS_STAT by HCP_STAT will have implications for the following
    parameter files besides SPVPD:

        CENSRACE will no longer be needed to assign race/ethnicity to the Census data
         since RACE will be included as a classification variable according to the process
         described above;

        CENSUNDER will no longer be needed to adjust for the Census undercount
         because the aforementioned process will include such an adjustment (although
         VA may want to retain this variable for flexibility in generating the database);
MEMO TO:   Allen Berkowitz and Stephen Meskin
FROM:      Jonathan Jacobson
DATE:      87/125/2003
PAGE:      8


       CHNGDEGR will need to indicate changes in degree of disability (if any) within
        HCP_STAT categories, and will be estimated from merged 3/31/2000,
        3/31/2001, 9/30/2001, and 9/30/2002 C & P mini master frozen files;

       CPDIS4 will no longer be needed because receipt of both compensation and
        pension benefits will no longer be reported within the model database;

       CPGENDER will no longer be needed because the CPNOMAT file will no longer
        be formed directly;

       DIS will need to be replaced by parameters indicating transitions between
        HCP_STAT values (this parameter file could be called CHNGHCP, and would
        need to vary by PERSVC since Gulf War veterans would never have HCP_STAT
        higher than 6P, and would also need to account for transitions into and out of the
        HCP_STAT categories defined by income rather than disability or period of
        service); merged 3/31/2000, 3/31/2001, 9/30/2001, and 9/30/2002 C & P mini
        master frozen files could used to estimate some of these transitions; longitudinal
        data from the 1996 panel of the Survey of Income and Program Participation
        (SIPP) that MPR already possesses could be used to estimate year-to-year
        transitions of veterans between the three income categories;

       DISADJ will no longer be needed to assign veterans to disability status because
        the aforementioned process will determine disability status within HCP_STAT;

       MIGRATE could be modified to differ according to some values of HCP_STAT
        as well as according to groups defined by AGE (although small cell sizes may
        create a need to combine HCP_STAT categories, even when separate categories
        are discernable in longitudinal C & P data);

       MORT could be modified to differ according to values of HCP_STAT, AGE, and
        GENDER instead of by values by DIS_STAT, AGE, GENDER, and
        officer/enlisted status;

       NEWDEGR will need to vary by HCP_STAT and will need to split HCP_STAT
        = 1AD by degree of disability (70, 80, 90, or 100), HCP_STAT = 1EF by degree
        of disability (50 or 60), HCP_STAT = 2GH by degree of disability (30 or 40),
        HCP_STAT = 3IJ by degree of disability (10 or 20), and HCP_STAT =
        3K_4L_5M by health care enrollment priority (3, 4, or 5); and

       PENSION will need to vary by HCP_STAT, AGE, and GENDER instead of by
        DIS_STAT, AGE, and GENDER.
MEMO TO:   Allen Berkowitz and Stephen Meskin
FROM:      Jonathan Jacobson
DATE:      87/125/2003
PAGE:      9



        HCP_6P will need to be created and vary by PERSVC to assign veterans with
         HCP_STAT = 6P_7Q or 6P_8R to health care enrollment priority 6.


   5. Next Steps

   Assuming this general approach is supported by VA, MPR will proceed to identify
   which programs will need to be changed, and will prepare written specifications to
   make the programming changes consistent with this conceptual design. In terms of
   changing parameters files, we recommend that priority be given to changing those files
   that must be changed in order to accommodate new classification variables such as
   HCP_STAT, as opposed to those parameter files for which changes are optional, even
   with a new database structure.


cc: Mark Brinkley (MPR)
    Chris Rankin (MPR)
    Natalie Justh (MPR)
    Mary Grider (MPR)
    Cathy Tomczak (VA)
    Floyd Watson (VA)
    George Sheldon (VA)
    Michael Wells (VA)
    Peter Ahn (VA)
    Robert Klein (VA)
    Eddie Thomas (VA)

								
To top