1. INTRODUCTION TO THE NONRESPONSE PROBLEM 
we also have to worry about whether a serious proportion of persons do not appear in the list. 
This might be the case for persons who do not live in private households (like inhabitants of 
hospitals or prisons) or who do not have official or permanent addresses (for instance 
homeless people). So in step 1 we are confronted with possible coverage errors, particularly 
the latter error of undercoverage. The noncoverage rate can sometimes be assessed. For 
example, in telephone surveys we could estimate it by looking up statistics on households 
with a telephone. There are also cases in which the amount of undercoverage error is not easy 
to measure, however, for example for the rate of homeless people or for persons residing in a 
country illegally. Other external information or comparative studies would be needed to do 
this. Kish (1965:531) writes that coverage errors are seldom reported. He suspects that there is 
a noncoverage rate"" of more than 10% for national samples. 
Step 2: 
Once a sampling frame has been accepted, one has to decide on a sampling design. As it is 
very often impossible to draw a simple random sample out of a complete list, a lot of 
strategies such as stratification, one-stage and multi-stage clustering or mixed designs have 
been developed. I will not discuss the different possibilities in detail because this topic 
constitutes a large separate field of knowledge. Explications of all the techniques are given by 
Kish" (1965, esp. chapters 3-6,10), Kalton (1983:8-56), and Verma (1998:1.1-1.40, 3.1-3.20, 
4.1-4.18). Each design has advantages and disadvantages and a compromise between data 
accuracy and survey costs has to be found. The aspects of costs are demonstrated by Groves 
(1989:49-80). Contrary to the situation in step 1, the amount of errors caused by the sampling 
design can be well assessed. There are formulas to calculate the so-called design effect 
Examples are given by Groves (1989:265-267, 271-279) and practical exercises by Verma 
(1998:1.14-1.21, 1.28-1.35). For this step, it has to be emphasised, that the sampling errors are 
in fact a known quantity". The magnitude can be predicted by our knowledge of the sampling 
design and probability theory. 
* One should keep in mind that since Kish's book in 1965, social changes in a society might have occured. An 
increase in social problem groups (e.g. homeless or criminal persons) could produce even higher noncoverage 
rates. 
"The complete book consists of more than 600 (!) pages dealing with sampling techniques. 
* Design effect is named "Deff", Kish (1965). 
For example, we can calculate weighting corrections for stratified and cluster samples. See: Kish (1965:77-82, 
148-166) or Kalton (1983:69-81). 
8