Tor Jakob Klette and Zvi Griliches, "The Inconsistency of Common Scale
Estimators when Output Prices are Unobserved and Endogenous", Journal of
Applied Econometrics, Vol. 11, No. 4, 1996, pp. 343-361.

OUR DATA SOURCES AND VARIABLE CONSTRUCTION:

The plant level data in our analysis are confidential. They can, however,
be made available on request to reseachers from Statistics Norway. The
full covariance matrix for all the variables needed to replicate our
analysis is documented in the ASCII file covmat.kg.

The data source used in our analysis is the annual census carried out
by Statistics Norway. Aggregate numbers and definitions for the census
are reported in NOS (several years). See also Halvorsen, R., R. Jenssen,
F. Foyn (1991) "Dokumentasjon av industristatistikkens tidsseriebase"
(Documentation of the manufacturing time-series files), Mimeo,
Statistics Norway.

Our unit of observation is a plant. We have employed an unbalanced
sample of annual observations for the period 1983-89 (inclusive) for 4
industry sectors: ``Metal products, except machinery and equipment''
(ISIC 381), ``Machinery'' (ISIC 382), ``Electrical apparatus and
supplies'' (ISIC 383), and ``Transport equipment'' (ISIC 384). The
sample includes only establishments with at least 5 employees. Plants
with incomplete reports for the variables needed in the estimation have
been eliminated, but no other cleaning has been carried out.

If we consider e.g. the year 1986, the number of plants vary from 196 in
``Electrical equipments'' (ISIC 383) to 663 in ``Metal products'' (ISIC
381), while the corresponding number of firms are 177 and 629
respectively. That is to say, almost all plants belong to separate
firms. The average number of observations per plant is about 4 in all
four industries. Summary statistics for a number of variables in the
employed sample are reported in table 1.

All variables are expressed in first differences of logarithms of the
variables. In some models, we have applied a Tornquist index for growth
in the variable inputs, i.e. the shares are constructed as the average
share for the two years used to construct the growth rates. Our
``output'' measure is nominal output deflated by the output deflator for
the industry. As discussed at length above, this is not a measure of
real output and it is the source of the biases we have emphasized
throughout this paper. Nominal output is adjusted for duties and
subsidies, and (in principle) also for changes in inventories in our
data. Price deflators for gross production (at seller prices),
materials, energy and capital (at buyer prices) are taken from the
Norwegian National Accounts. Labour inputs are represented by man hours.
Wage payments comprise salaries and wages in cash and kind, other
benefits for the employees, taxes and social expenses levied by law.
Energy use and material inputs are reported as separate variables in our
data set.

The capital input variable employed is based on investment figures and
the total reported fire insurance value for buildings and machinery. The
annual movements are obtained by assuming geometric depreciation at a 3
percent annual rate, and that investment takes about a year to become
productive. An examination of the fire insurance values and a comparison
with the investment figures reveal much noise in the fire insurance
values. We have therefore constructed a simple filter to pool the two
sources of information about movements in the capital stock. We pool all
the information in the fire insurance values and the investment figures,
using the perpetual inventory method, to estimate the level of the
capital stock for the first observation for each plant. Extreme fire
insurance values have been eliminated. From this initial level, the
capital stocks in the subsequent years are estimated by the perpetual
inventory method.

The growth in industry output is estimated by according to the formula
presented in section 4.2. We have used the share of industry sales as
weights. More precisely, when estimating the growth in industry output
from year t-1 to t, we have weighted together the growth in deflated
sales for all plans in the industry, with the average output shares for
the plant in the two years as weights.

There are two ASCII files. The file covmat.kg is designed to be easy to 
read. The file covacc.kg is harder to read, but the numbers have far more 
significant digits. Both ASCII files are zipped in the file kg-data.zip.

VARIABLES IN THE COVARIANCE MATRICES IN THE ASCII-FILE:

    gQ : growth in output
g(M-L) : growth in materials per unit of labor
g(E-L) : growth in energy per unit of labor
    gK : growth in capital
    gL : growth in labor (man hours)
    gX : growth in (cost) share weigheted variable inputs
 gQind : growth in industry output
    gN : growth in number of employees