You are on page 1of 5

Index Number: 090306

Title: Controlling the Distribution of Response Burden in


Longitudinal and Cross Sectional Business Surveys

Author: Ismo Teikari

Date:

Country: Finland

Round Table: 9th Round Table

Örebro (SWEDEN)

September 18 - 22, 1995


Session 3.

Controlling the distribution of response burden in longitudinal


and cross sectional business surveys
Ismo Teikari, Business Register

Increasing interest to survey small and medium size enterprises (SME) has brought
up the problem of response burden of businesses. The reduction of response burden in
systems of statistical agents could be implemented by comprehensive planning of
business surveys carried out by the statistical agent.

Even the more important task is to control the distribution of response burden. When
drawing numbers of business samples we burden randomly businesses. This means that
the distribution of response burden is uneven. Solution to this problem may be reached
by using such sampling designs that make it possible to control response burden of
businesses.

The beginning of history could be placed to 1962 when the article "Development of
sampling plans by using sequential (item by item) selection techniques and digital
computers" by Fan, Muller and Rezucha was published in American Statistical
Association Journal. They presented and gave proof for the sequential simple random
sample which later were implemented as list sequential simple random sampling in the
SAMU system in Statistics Sweden.

Using permanent random numbers and ordering businesses according these numbers
it is possible to control how enquiries burden businesses. This is very simple when using
simple random sampling (SRS) designs. Unfortunately the distribution of business
population is so skew that the totals have tendency to be underestimated if the size of
businesses is not taken into account in sample inclusion probabilities.

The problem of sampling with probability proportional to size has a history that goes
back to decade 1940. Brewer and Hanif (1983) have made a comparison between 50
methods that have developed since then. None of these give the perfect method for the
problem. Most of the methods are usable only if the sample size is not over two. One of
the most usable methods is the Poisson sampling. It is very usable in sample coordination
but its drawback is the random sample size. On the other hand the random sample size
may be a minor problem compared to nonresponse for example. Some modified
procedures have been developed but they have the tendency to distort either positive or
negative coordination.
The control of sampling surveys must work both in cross sectional survey samplings
and in longitudinal samples. Positive coordination means control over longitudinal
samples so that some units are included in successive samples. Longitudinal surveys give
better estimates than cross sectional surveys because it follows the same units over time.
The negative coordination means control over simultaneous or successive samples so that
the inclusion of same units is minimal. By using list sequential SRS it is easy to manage
both positive and negative coordination. When using stratified sampling design the
problem arises from shifts of units between different strata in successive surveys. In the
Dutch system this is solved by sorting the data according to rate of response burden in
ascending order. After sorting the businesses which have the least cumulative rate of
response burden are located in the beginning of each strata. In the French system the
same problem is solved by changing the random number after each draw so that the units
belonging to the last sample move to the end of each strata.

In Finland we will use the data where units are ordered according to permanent
random numbers and examine sequentially (item by item) the cumulative response
burden. Before searching the solution to the above-mentioned problem concerning the
shifts between strata, we will study if it is only a minor problem. We will make this
examination during the autumn in panel data which is constructed for the business
demographic study.

We will set the focus on the probability proportional to size sampling. The most
suitable one seems to be the Poisson sampling. The rotation of Poisson samplings
proceeds as follows (picture 1). In the first sample those units are drawn which lie above
the line ac. In the next sample those units are drawn which lie above the line beginning
from point a'. It contains also the largest units in interval aa'. The other units in the
interval aa' are released from rotation until the next round.

Using the Poisson rotation there is an area aa'b (Picture 1) which includes the units
never drawn in longitudinal samples. When we shift the starting point from a to a' the
units in the area (aa'b) are not drawn in either sample.

The solution (picture 2) we have constructed is as follows. We will rotate the area
abed where the area abcd is the list sequential SRS and the area cde is the list sequential
Poisson sampling. The draw of units into the sample happens as follows. The unit is
accepted in the sample if it lies in the area abed. When we draw the next sample we
move the starting point to point d. The units lying in the area abcd are set free from
panel except the biggest ones. The most of small enterprises are only once in the panel
until next rotation round.
PICTURE 1 ROTATED POISSON SAMPLING

Number of employees
c
400

300

200

100 b

0
0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1
a a'
Permanent random number

PICTURE 2 ROTATED SRS + POISSON SAMPLING

Number of employees
b c e
400

300

200

100

0
0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1
a d
Permanent random number
References

Brewer K.R.W., Hanif M. Sampling with Unequal


Probabilities. Lecture Notes in Statistics 15. New
York. 1983

Cotton, F., Hesse, C., Co-ordinated salection of stratified


samples. Proceedings of Statistics Canada Symposium 92,
Design and Analysis of Longitudinal Surveys, November
1992.
Fan, C.T., Muller, M.E., Rezucha,I. (1962). Developement of
sampling plans by using sequential (item by item) selection
techniques and digital computers. J. Amer. Statist. Ass.
57, 387-402.

Koeijers, E. (1993) EDS: Sampling System for the Central


general Business Register-A Method to Spread Response
Burden. First Eurostat International Workshop on
Techniques of Enterprise Panels
Ohlsson E. (1992). SAMU (The System for Co-ordination of
Samples from the Business Register at Statistics
Sweden). Unpublished Paper.

You might also like