Maxim Engers, Monica Hartmann, and Steven Stern, "Mileage Drives Used 
Car Prices", Journal of Applied Econometrics, Vol. 24, No. 1, 2009,
pp. 1-33.

This paper has a number of files associated with it. All data files are
zipped in ems-data.zip. All program files are zipped in ems-programs.zip.
Several Excel spreadsheet files are zipped in ems-xls.zip.

All ASCII files are in DOS format. Unix users should use "unzip -a". Note
that file sizes given below refer to files in Unix format.

ehs-programs.zip:
  dataextract.f
  nptsestimate.f

ehs-data.zip:
  npts.d
  data_mileage_drives_prices.prn
  data_mileage_drives_prices.txt
  olsdata.prn
  nptsestimate.p
  
ehs-xls:
  mileageprice.xls
  mileagepriceb.xls
  mileagepricec.xls

There are a number of data sources used in this paper. These are described
below:

1) We use an extract of the 1995 National Personal Transportation Survey
(1995).  The FORTRAN program dataextract.f (1072 lines of code) can be used
to replicate our data.  The original data and full documentation is
available publicly through, for example, ICPSR.  A list of variables pulled
from the NPTS is described below:

NPTS Household record:
Name of Variable in FORTRAN program | Description | FORTRAN format

ivar(1) availability of public transportation 22x,i2
ivar(2) household id # 14x,i8
ivar(3) household income 8x,i2
ivar(4) year of interview 44x,i4
ivar(5) availability of public transportations 6x,i2
ivar(6) member age i4
ivar(7) whether member is a driver i2
ivar(8) whether member is male 2x,i2
ivar(9) whether member works 2x,i2
ivar(10) member age i4
ivar(11) whether member is a driver i2
ivar(12) whether member is male 2x,i2
ivar(13) whether member works 2x,i2
ivar(14) member age i4
ivar(15) whether member is a driver i2
ivar(16) whether member is male 2x,i2
ivar(17) whether member works 2x,i2
ivar(18) member age i4
ivar(19) whether member is a driver i2
ivar(20) whether member is male 2x,i2
ivar(21) whether member works 2x,i2
ivar(22) member age i4
ivar(23) whether member is a driver i2
ivar(24) whether member is male 4x,i2
ivar(25) whether member works 2x,i2
ivar(26) member age i4
ivar(27) whether member is a driver i2
ivar(28) whether member is male 2x,i2
ivar(29) whether member works 2x,i2
ivar(30) member age i4
ivar(31) whether member is a driver i2
ivar(32) whether member is male 2x,i2
ivar(33) whether member works 2x,i2
ivar(34) member age i4
ivar(35) whether member is a driver i2
ivar(36) whether member is male 2x,i2
ivar(37) whether member works 2x,i2
ivar(38) member age i4
ivar(39) whether member is a driver i2
ivar(40) whether member is male 2x,i2
ivar(41) whether member works 2x,i2
ivar(42) member age i4
ivar(43) whether member is a driver i2
ivar(44) whether member is male 2x,i2
ivar(45) whether member works 2x,i2
ivar(46) (not used in analysis) 125x,i4 weight household weight f10.2
ivar(47) urban 4x,i2 aivar alphanumeric urban 140x,a1


NPTS Vehicle record:
Name of Variable in FORTRAN program | Description | FORTRAN format

jveh(1) household id # i8
jveh(2) miles driven / 10000 30x,i6
annm mod(miles driven,10000) f8.2
jveh(3) make/model information 6x,i2
jveh(4) make/model information 2x,i3
jveh(5) make/model information 14x,i2
jveh(6) model year i4
jveh(7) month of odometer reading 82x,i4
jveh(8) year of odometer reading i4
jveh(9) measure of miles driven 12,i6


Make and model information is aggregated in a FORTRAN subroutine called
maketr.  It transforms brands according to the Brand Transformation table.


Brand transformation:
Aggregated Model (Name, Number) | Make [ivar(3)] | Model [ivar(4)] |
Whether vehicle [ivar(5)]

not a vehicle | 26 | | | 0
isuzu | 1 | 38 | | 1
chrysler | 2 | 6 | | 1
dodge | 3 | 7 | | 1
plymouth | 4 | 9 | | 1
ford | 5 | 12 | | 1
mercury | 6 | 14 | | 1
chevrolet | 8 | 20 | not 34,35 | 1
geo | 23 | 20 | 34,35 | 1
oldsmobile | 9 | 21 | | 1
pontiac | 10 | 22 | | 1
saturn | 11 | 24 | | 1
american luxury | 12 | 13,19 | | 1
japanese luxury | 13 | 58,59 | | 1
european luxury | 14 |32,34,39,42,45,47 | | 1
honda | 15 | 37,54 | | 1
mitsubishi | 16 | 52 | | 1
mazda | 17 | 41 | | 1
nissan | 18 | 35 | | 1
subaru | 19 | 48 | | 1
toyota | 20 | 49 | | 1
volkswagen | 21 | 30 | | 1
volvo | 22 | 51 | | 1
hyundai | 24 | 61 | | 1
other | 25 | all others | 1


2) The data extract from the 1995 NPTS described above is stored in a
hierarchical ASCII file called npts.d with 210097 records requiring 5411115
bytes.  There are three types of records in the file.  For each household,
there is a "household record." Then it is followed by "household member
records" for each household member, and then "vehicle records" for each
vehicle owned by the household.  The code necessary to read the data is in
the subroutine datget in the FORTRAN program nptsestimate.f (7227 lines of
code).  The structure of each type of record is:


Household record:
Name of Variable in Fortran Program | Description | FORTRAN format

ihouseid | household id # |1x,i8
weight | household weight | 1x,f8.2
ihousinc | household income | 1x,i5
ibusavl | (not used in analysis) | 1x,i2
ipbtravel | (not used in analysis) | 1x,i2
iurban1 | urban dummy | 1x,i2
iurban2 | (not used in analysis) | 1x,i2
nhchar | # household members | 1x,i2
ihco | # household cars | 1x,i2
ihcar | (not used in analysis) | 1x,i2


Household member record:
Name of Variable in Fortran Program | Description | FORTRAN format

jt | (not used in analysis) | 1x,i2
ihchar(1) | whether driver | 1x,i2
ihchar(2) | whether adult driver | 1x,i1
ihchar(3) | whether male | 1x,i1
ihchar(4) | whether drives to work | 1x,i1


Vehicle record:
Name of Variable in Fortran Program | Description | FORTRAN format

mveh(1) | annual miles driven | 1x,i6
mveh(2) | brand | 1x,i2
mveh(3) | model year | 1x,i4
mveh(4) | (not used in analysis) | 1x,i4
mveh(5) | (not used in analysis) | 1x,i4
mveh(6) | (not used in analysis) | 1x,i6

3) We use data collected from Kelley Blue Book's Official Guide for Older
Cars (May/June - Central/Eastern Edition, 1995-2000) and the 1995 NPTS to
run regressions of price on miles driven as a function of age. The Kelley
Blue Book variables used in our study are described in the ASCII file
variables_mileage_drives_prices.txt, and the data are saved in the
space delimitated ASCII file data_mileage_drives_prices.prn.  The data from
data_mileage_drives_prices.prn is manipulated into a form directly used in
the OLS regression and can be read from a rectangular ASCII file called
olsdata.prn with 26 records.  The code necessary to read the data is in the
subroutine decompos1 in the FORTRAN program nptsestimate.f (7227 lines of
code).  There are two variables per record; the first is log(avg miles by
brand), and the second is log(average price by brand).  The order of the
brands is in the FORTRAN statement:

      data abrand/'isuzu','chrysler','dodge','plymouth','ford',
     & 'mercury','buick','chevrolt','oldsmobl','pontiac','saturn',
     & 'luxamer','luxjapan','luxeurop','honda','mitsubi','mazda',
     & 'nissan','subaru','toyota','volkswg','volvo','geo','hyundai',
     & 'other','truck'/


Other data are in a set of excel files called mileageprice.xls (54 KB),
mileagepricea.xls (38 KB), and mileagepriceb.xls (38 KB).  The underlying
data are also available in the text file data_mileage_drives_prices.prn, but
not any of the analysis.


4) A file with initial values of parameters is in nptsestimate.p (212
records) and is read in the subroutine parget in nptsestimate.f (7227 lines
of code). The order of variables in the file is:

  method,nobs,nmodel,ncchar,nhouse,ncarmax,npoly,ishare,itotmil
  coef(alpha,beta,delta,gamma,omega,sigu)
  icoef
  alabel

alabel  = labels for coefficients
alpha   = derivatives of v wrt c
beta    = coefficients on products of polynomials
coef    = initial parameter values
delta   = polynomial coefficients
gamma   = derivative of log h wrt household characteristics
icoef   = flags for restriction
ishare  = flag for using shares in estimation
itotmil = flag for using total miles in estimation
method  = 0 for evaluation of log likelihood function
          1 for estimation
          2 for alternative estimation method
          3 for estimation with amoeba
          4 for covariance estimation       
          5 for ols estimation
          6 for mom logit
          7 for gridsearch
          8 for computing age effects
          9 for computing age derivatives
         10 for computing demographic derivatives
         11 for decomposing derivatives
         12 for decomposing derivatives using alternative mehtod
         13 for preparing data for cex analysis
         14 for plotting miles
         15 for simulation
ncarmax = max k for variation in omega
ncchar  = # car characteristics
nhouse  = # household characteristics
nmodel  = # models
nobs    = # observations
npoly   = order of polynomial
omega   = derivative of psi wrt v(ranked)
sigu    = standard deviation of u (not identified)


5) nptsestimate.f uses a simplex method to optimize from a commercial
product called GQOPT.  All calls to subroutine opt in nptsestimate.f are
calls to use the subroutine NMSIMP in GQOPT.  The user should replace this
code with an alternative optimizer available to him/her.  The name of the
function being optimized is the 9th argument in the call to opt.  

Steven Stern
Sns5r [AT] virginia.edu