561
Non-Response in Probability Sample Surveys
in the Czech Republic*
JINDŘICH KREJČÍ**
Institute of Sociology, Academy of Sciences of the Czech Republic, Prague
Abstract: In this article the problem of survey non-response is examined with
special reference to probability sampling in the Czech Republic. Non-response
rates among Czech respondents in ISSP surveys between 1995 and 2005 were
almost twice the rate recorded between 1991 and 1995 (25%). Such trends
point to a decline in the `survey climate'. While non-contacts and refusals in
surveys are a significant problem, issues relating to how fieldwork is undertaken
are equally important. The large fluctuations in non-contact rates and
the relative success of the Czech Statistical Office in attenuating non-response
rates demonstrates that prudent surveying strategies can be effective. An examination
of two waves of the European Social Survey (ESS) reveals both the
problems and potential strategies available for response rate enhancement. In
this respect, all survey designers face the dilemma of balancing the benefits
of data accuracy with increasing logistical costs. Improvement in survey quality
necessitates consideration of many issues and the ability to make sensible
trade-offs between competing research objectives.
Keywords: survey research in the Czech Republic, response rates, non-response
trends, fieldwork strategy
Sociologický časopis/Czech Sociological Review, 2007, Vol. 43, No. 3: 561­587
Introduction
This article deals with the issue of non-response rates in social survey research in
the Czech Republic. It examines current trends and provides an in-depth analysis
of this issue using the first two waves of the European Social Survey (ESS) in the
Czech Republic.
The capacity of any quantitative analysis to provide evidence about social
reality depends largely on the quality of the data gathered. Mass surveys, based
* The study was prepared as a part of the Grant Project No. 1N04192 supported by the
Ministry of Education, Youth and Sports of the Czech Republic, and the Grant Project No.
403/06/1421 supported by the Czech Science Foundation. The author would like to thank
Pat Lyons for very useful recommendations and his help with the English version of the
paper and two anonymous reviewers for their valuable comments.
** Direct all correspondence to: Jindřich Krejčí, Institute of Sociology, Academy of Sciences
of the Czech Republic, Jilská 1, 110 00 Prague 1, Czech Republic, e-mail: jindrich.krejci@soc.
cas.cz.
socrev2007-1.indb 561socrev2007-1.indb 561 3.9.2007 18:41:303.9.2007 18:41:30
Sociologický časopis/Czech Sociological Review, 2007, Vol. 43, No. 3
562
on interviewing population samples, are among the most commonly used sources
of data, and the quality of their analysis derives directly from how well these surveys
represent their target populations. The level of representativeness depends
on a number of factors. Key concerns are the level of coverage of the target population,
the size of the sample, and the appropriateness of the sampling method
used. Moreover, the degree to which researchers are able to reach and interview
the selected respondents is of fundamental importance. Consequently, survey response
rates are one of the key indicators of the quality of all survey data.
There are various ways of monitoring response rates, but essentially it is a
matter of the ratio of the number of completed survey questionnaires received
back and processed to the number of questionnaires distributed to eligible addresses
in the sample. Complementary survey non-response is the failure to obtain
data from the entire sample selected for interview. There are two main types
of survey non-response. First, there is `item non-response', where some respondents
do not answer specific questions in an interview. Second, there is `unit nonresponse',
where specific individuals (or households) selected for an interview
do not take part in the survey. This article will focus on unit non-response, as this
is the most salient problem for all survey research.
The majority of surveys involve obtaining the voluntary cooperation of respondents.
However, practical circumstances (e.g. the unavailability of the appropriate
sampling frames, regulations pertaining to the protection of personal data,
etc.) in many countries require the application of complicated sample designs.
In addition different groups in the population maintain lifestyles that make it
difficult to contact their members. There are also a number of organisational factors
that can result in the failure to interview selected respondents. Therefore,
achieving a satisfactory response rate can pose quite a problem. The successful
processing of all the distributed questionnaires, that is, achieving a hundred-percent
response rate, practically never occurs.
Based on experiences the prevailing wisdom among researchers is that
since the mid-1990s there has been a dramatic decline in the response rates of
sample surveys conducted in the Czech Republic. This view gives rise to serious
concern about the future prospects of Czech social research.1
However, little
scholarly attention has yet been devoted to this issue in the Czech Republic.
Detailed examinations of response rates have been the subject of some reports
prepared within the framework of individual research projects [e.g. Tuček, Gatnar
and Krejčí 2000], but such texts, prepared for survey subscribers, are not often
available to the wider public. There is also a book-length examination of the
quality of political survey's estimates of voter preferences that draws attention to
the issue of response rates [Krejčí et al. 2004]. However, to date there has been no
1
This issue was discussed, for example, in the panel discussions `What Methodology for
Czech Sociology?' and `Research Infrastructures' at the conference `Czech-Slovak Sociology
Days' in Prague, 10­12 May 2005. http://web.soc.cas.cz/cssd/.
socrev2007-1.indb 562socrev2007-1.indb 562 3.9.2007 18:41:303.9.2007 18:41:30
Jindřich Krejčí: Non-Response in Probability Sample Surveys in the Czech Republic
563
comprehensive study of the nature and potential impact of survey non-response
on the quality of survey data generated within the Czech Republic.
The problems evident in Czech survey research do not always confirm the
prevailing opinion of continual decline. There are also surveys with relatively
high response rates, and, for example, the two Czech surveys conducted as part
of the research programme of the European Social Survey (ESS) achieved substantially
different outcomes. In the survey conducted in 2002­2003 the response
rate was 43.5%, while in 2004 the response rate during the fieldwork was 71.0%.
The ESS is an international academic research programme that espouses
strict methodological standards. Both waves of this survey conducted in the
Czech Republic adhered to most of the methodological requirements stipulated
by the ESS, but they differed in the emphasis put on applying procedures aimed
at increasing response rates. In the context of the problems surrounding response
rates, it is important to examine what factors may have contributed to generating
such varied outcomes. However, owing to the marginal amount of attention
devoted to the issue of response rates in the Czech Republic it is first necessary to
look at the issue more generally and describe the current Czech situation.
Consequently, this article will focus on three key themes. In the first section,
there will be a general overview of the issue of low response rates in survey research.
This is followed by an examination of the non-response trend in the Czech
Republic. The third section discusses differences in response rates in the first and
second waves of the European Social Survey in the Czech Republic. In the final
section some concluding remarks are made regarding the evidence presented.
The problem of non-response in sample surveys
A low response rate is often regarded as one of the main sources of variable variance
and bias in mass survey estimates [see, e.g., Särndal and Lundström 2005;
Dillman et al. 2002; Biemer and Lyberg 2003], though the linkage between nonresponse
rates and survey errors is not direct and may even be absent [see Groves
2006]. There is a consensus among scholars and pollsters that response-rate trends
have been declining, and that this fact poses one of the most serious problems affecting
contemporary quantitative social science research [e.g., Bradburn 1992;
Bogart 1987; Särndal and Lundström 2005: 11­15].
Two main reasons are most often cited as the cause of this decline in response
rates. The first reason paradoxically stems from the boom in the research
industry. The number of surveys undertaken each year has been increasing for
decades. According to estimates in advanced industrial countries the majority of
potential respondents have been surveyed, often several times, and the population
could be described as `over-surveyed'. People moreover have difficulty distinguishing
between surveys and direct marketing, which competes with survey
research by infringing on the privacy of respondents. The result is an overall insocrev2007-1.indb
563socrev2007-1.indb 563 3.9.2007 18:41:303.9.2007 18:41:30
Sociologický časopis/Czech Sociological Review, 2007, Vol. 43, No. 3
564
crease in refusals. The second reason relates to changes in lifestyle, which make
it increasingly more difficult to reach the respondents selected for interview and
therefore cause an increase in non-contact rates.
Non-response trends
Steeh [1981] reports a large increase in the number of refusals in survey research
in the United States between 1952 and 1979. This decline in response rates has
subsequently been confirmed in numerous analyses focusing on different survey
programmes in various fields of research and in individual countries. However, it
is not easy to document general trends because there are big differences between
survey methods and their consequences, and the situation varies also between
countries and different research fields and topics. There have been few long-term
research programmes based on a stable methodology that would enable a deeper
evaluation of this phenomenon. Those that do exist do not encompass all fields of
research and moreover their analyses have produced ambiguous results.
Several analyses of trends in response rates over time and cross-nationally
have confirmed the existence of a general decline [Baim 1991; de Leeuw and de
Heer 2002]. While other studies do not dispute the fall in response rates in the
past, they do not confirm the continuity of the decline [de Heer 1999; Smith 1995;
Steeh et al. 2001; Groves and Couper 1998: 156­172]. According to practitioners
from survey agencies [e.g. Sheatsley 1987] it is still possible to attain high response
rates, but it requires more effort and higher costs in organising the research. This
observation can be readily seen in the outcomes of surveys that emphasised the
use of a rigorous data quality methodology, for example, surveys conducted by
statistical offices or the General Social Survey in the United States [de Heer 1999;
Groves 2004: 145­153; Smith 1995]. These studies regularly achieve a response
rate of at least 70% in countries where a significant fall in response rates has been
registered.
However, there tends to be enormous differences between outcomes when
different methodologies are used. The Council for Marketing and Opinion Research
(CMOR) conducts the `Respondent Cooperation and Industry Image
Study' and monitors the response rates of many different surveys.2
Among other
things, its studies reveal that the response rates of commercial telephone surveys
conducted in the United States are often below 25%. Steeh et al. [2001] studied the
response rate of commercial surveys conducted using the `random digit dialling'
(RDD) method. According to their findings, in the 1960s and 1970s there was a
sharp increase in non-response and especially in refusal rates. This trend subsided
in the 1980s and 1990s, and now survey response rates appear to be stable.
However, the nature of non-response has also been changing, to the detriment of
measurement accuracy, owing to an increase in overall non-contact rates and a
decrease in refusals, especially in metropolitan areas, and other reasons.
2
See the research section of the CMOR website: http://www.cmor.org/.
socrev2007-1.indb 564socrev2007-1.indb 564 3.9.2007 18:41:303.9.2007 18:41:30
Jindřich Krejčí: Non-Response in Probability Sample Surveys in the Czech Republic
565
Inference in surveys with non-response and treatment of the problem
From the perspective of statistical inference, survey non-response has an impact
on point estimates in terms of bias and the inflation of variance and bias in the
estimators of precision [Dillman et al. 2002: 4­6; Särndal and Lundström 2005;
Groves 2004: 158­159]. However, it must be stressed that estimations of error
caused by non-response are theoretical, and any exact assessment of the impact
of lower response rates is usually problematic. This is because the non-response
error does not stem from the non-response rate as such, but is the result of the
extent of the differences between the attitudes and characteristics of respondents
and non-respondents with respect to the measured phenomenon. At the same
time, relevant data on non-respondents are by definition missing. Groves [2004:
133] summarises this approach with the following equation.
RESPONDENT VALUE =
FULL SAMPLE VALUE + (NON-RESPONSE RATE) x
(RESPONDENT VALUE ­ NONRESPONDENT VALUE)
If the attitudes of respondents and non-respondents are the same then a
low response rate does not matter. Several empirical studies confirm this fact, describing
situations when higher non-response did not lead to higher biases [e.g.
Keeter et al. 2000; Merkle and Edelman 2002: 253­255; Groves 2006]. However,
many analyses have shown that response rates are correlated with population
characteristics and thus also associated with biases of estimations. On the other
hand, the impact of these correlations does not usually invalidate the results of a
survey and can be often corrected by weighting [e.g., Stoop, Iedema and Louwen
2000; Keeter et al. 2000]. These relationships also depend heavily on the topic
of research [e.g. Zaletel and Vehovar 1998], and they do not affect all variables
equally [e.g. Joye 2004], so the impact of non-response on survey data quality
cannot be generalised.
While non-response error cannot be ignored, it is often not possible to make
any reliable estimate of its size. Although many researchers consider the sample
survey method to be relatively robust, non-response has the potential to be
the source of some unpleasant surprises. One of the consequences of this lack of
transparency is that non-response error is not usually taken into account in survey
estimates and statistical tests. However, a low response rate is often regarded
as sufficient reason for commissioning agencies, researchers, journal editors, and
other users of data to question the accuracy of survey research results. This is
reflected in the substantial effort invested in developing and applying methods
aimed at increasing response rates and reducing non-response error.
Statistics offer two basic approaches to addressing non-response: imputation
and weighting. Imputation means substituting the values for missing units
with estimates developed on the basis of auxiliary information obtained in the
research or from other sources. But in the case of unit non-response this is often
not feasible. The most frequently used weighting procedures are based on comsocrev2007-1.indb
565socrev2007-1.indb 565 3.9.2007 18:41:313.9.2007 18:41:31
Sociologický časopis/Czech Sociological Review, 2007, Vol. 43, No. 3
566
paring the socio-demographic characteristics of surveys with those published in
official statistical sources. On the basis of observed differences weights are assigned
to cases in the data file with the objective of matching the profile of the
sample dataset with the known composition of the target population. However,
Joye [2004] has shown that other variables, such as participation and social integration,
which have only an indirect relationship to the `objective' criteria normally
used in weighting, may also be strongly correlated with the propensity
toward non-response.
Post-fieldwork adjustments of survey data can reduce the level of error, but
such procedures cannot fully make up for data that are missing. However, there
are a number of methods used to improve response rates that are applied in the
survey preparation phase and in the course of data collection. Among the most
frequently used strategies [Biemer and Lyberg 2003: 91­115] are advance letters of
introduction, providing incentives to respondents, increasing the number of contact
attempts and call-backs, adopting techniques of refusal conversion, adapting
the timing of fieldwork to match respondents' lifestyles, constructing questionnaires
with due regard to respondent burden and likely interest in the content of
the survey interview, improving interviewer selection and training, and utilising
mixed modes of interviewing, etc.
It should be always borne in mind that the objective here is not only to
increase the response rate but also and primarily to reduce non-response error.
Some methods can help increase response rates, but they can also have the unintended
consequence of increasing the level of bias in a survey. Techniques of
refusal conversion, for example, do not help if the main source of bias is the low
representation of hard to reach groups within the population. Similarly, incentives
to encourage respondents to participate must be applied cautiously, because
different groups of the population respond to `prodding' in different ways. It
should also be remembered that the key issue of survey data quality is efficiency.
In short, efforts devoted to ensuring survey accuracy need to be tempered by
cost considerations. Therefore, careful thought needs to be given to the kinds of
remedial methods implemented.
The response rates are affected by a number of mutually inter-connected
factors. Some of them, such as the social context and the respondent's lifestyle, are
beyond the researcher's control, but they are important sources of the potential
biases caused by non-response. Addressing other factors, such as the performance
of interviewers, is not just a task that lies within the scope of how a single survey is
organised, but is often the result of long-term systematic work. Successful strategies
for reducing non-response error are therefore usually based on the combined
use of various methods, the selection of which depends on external circumstances.
Fortunately, there is currently a substantial amount of literature available on the
effects associated with the use of various surveying methods, and theories on survey
participation [e.g. Groves and Couper 1998] provide the necessary conceptual
foundations for assessing the strategies aimed at reducing non-response.
socrev2007-1.indb 566socrev2007-1.indb 566 3.9.2007 18:41:313.9.2007 18:41:31
Jindřich Krejčí: Non-Response in Probability Sample Surveys in the Czech Republic
567
Non-response trends in probability surveys in the Czech Republic
Before the Velvet Revolution in 1989 very specific conditions existed in Czechoslovakia
for conducting sample surveys. Social research was controlled by the
totalitarian regime. As a result, relatively few surveys were conducted and their
thematic focus was restricted. Interaction with respondents took place in an atmosphere
of suspicion, where overt expressions of loyalty to the regime were
obligatory. Given these special circumstances it is not surprising that the decline
in response rates described in many Western countries was not witnessed in
Czechoslovakia during the 1970s and 1980s.
Survey climate
After 1989 survey research underwent a fundamental change for a number of reasons.
The number of public opinion surveys grew substantially, and a new and
large sector of market research emerged. Moreover, the country became involved
in international academic research programmes. At the same time the external
conditions in which research has been conducted changed in distinct ways [see
Krejčí 2004].
First, at the end of 1992 Czechoslovakia was divided into two separate states,
the Czech Republic and Slovakia. This led also to the gradual division of research
structures. Second, probability sampling has become significantly more complicated.
Since 1993 researchers have not had access to the Central Population Register.
Since 2001 the database of households, which is maintained by the Czech
Statistical Office, has only been open to researchers working on official statistical
surveys. The quality of samples based on the commercial database of addresses
from SIPO, a service through which the public can make regular bill payments, is
now questionable given the widespread availability of alternative means of making
payments. These constraints resulted in a lack of quality sampling frames,
which complicates the design of samples and increases costs. Third, since 2002
strict legislation has applied to the protection of personal data. This further complicates
the process of sampling, collecting, and processing data.
However, not all changes to the survey climate have been institutional in
nature. Lifestyle changes stemming from the political and economic transition
process have also had an important impact, not only on the topics examined in
survey research, but also on the manner in which the research is undertaken.
One example of this process is that some population sub-groups have become
harder to contact. Rising crime rates have made it more difficult for interviewers
to establish contact with increasingly wary respondents. The main entrances of
residential buildings now tend to be locked and the initial moment of contact
occurs through an intercom. It is also harder to gain the public's trust. The rise
in the number of marketing activities has contributed to a greater emphasis on
protecting the privacy of individuals.
socrev2007-1.indb 567socrev2007-1.indb 567 3.9.2007 18:41:313.9.2007 18:41:31
Sociologický časopis/Czech Sociological Review, 2007, Vol. 43, No. 3
568
Negative experiences from the totalitarian past have resulted in a prevailing
fear about the abuse of personal information on individuals and households. For
example, the Czech survey on Social Stratification in Eastern Europe after 1989
became the target of a massive assault from the media. During the course of data
collection in 1993 researchers were accused of preparing a database of information
on individuals for use in an anti-reform plot. This led to a decline in the response
rate of this survey. Even official statistical surveys such as the 2001 Census
have been accused of misusing information, and the need to collect such data at
all has been called into question.
Survey non-response
These changes have been accompanied by unfavourable developments in response
rates. Despite the lack of any systematic research on outcome rates the
prevailing wisdom within the Czech survey-research community in this respect
may be summarised as follows:
(a) In the mid-1990s there was a sharp increase in non-response rates in Czech
surveys.
(b) In the following decade this trend accelerated.
Unfortunately, it is not possible to make an extensive and in-depth analysis
of the trends for three main reasons. First, the examined period is not long,
because social research has only been developing freely in the Czech Republic
for a short time ­ less than two decades. Second, there are few publicly accessible
probability surveys that allow a detailed monitoring of non-response rates.
Many important surveys have been based on quota samples and do not provide
relevant data on response rates.3
Third, the circumstances described above also
mean that there is no long-term survey that has not been affected by significant
changes in its methodology, and thus identifying the causes of non-response can
be difficult. Nevertheless, there is still enough evidence to evaluate the validity of
the above-mentioned conviction generally shared by experts.
Standards for publishing information on response rates have not been generally
established in the Czech Republic. Consequently, such information is not
3
The number of refusals can be reported also for quota samples, but it provides only limited
information about the non-response problem and is incomparable with outcomes of
probability surveys. At the same time, in quota surveys interviewers are less motivated to
interview hard to reach and less compliant members of the population, and the potential
for error is consequently even greater. The higher level of obscurity of the non-response
problem in quota surveys and its possible risks is one of the reasons why quota samples
are a less preferred survey method [e.g. Kaase 1999]. On the other hand, for example,
Groves shows that the problem is more complicated and with respect to non-response
in both probability and non-probability sample surveys, researchers are required to deal
with the same set of methodological questions, but they have different resources to answer
them [Groves 2006: 667­668].
socrev2007-1.indb 568socrev2007-1.indb 568 3.9.2007 18:41:313.9.2007 18:41:31
Jindřich Krejčí: Non-Response in Probability Sample Surveys in the Czech Republic
569
available for a number of surveys. In other cases response rates are overestimated
because not all types of non-response are included. For example, there is a tendency
to ignore interviewer failures, and at some agencies these are the source of
as much as one-third of all non-responses. Sometimes only refusals are taken into
account, so what is really being reported is the `cooperation rate'. If such information
is used in the place of the total response rates, which happens regularly,
the result is an underestimation of the risk of biased inferences resulting from
non-response error. However, concealing or providing incomplete information
on response rates is an infringement of professional standards and against the
rules of the international associations such as WAPOR, ESOMAR and EFAMRO
[see Smith 2002], and SIMAR, the professional market research association in the
Czech Republic.4
Therefore, only surveys that are publicly accessible and whose metadata
include a detailed breakdown of outcomes allowing a comparison of response
rates have been used. Calculations of survey response and non-response rates are
based on definitions specified by AAPOR [2006]. All types of non-response are
taken into account, except incorrect or unoccupied addresses and administrative
errors connected with the sampling process.
Figure 1 presents non-response trends in the area of academic research.
These are Czech surveys conducted as part of international research programmes,
and I have used them because they provide all the necessary information for
analysis. All surveys are based on probability sampling of individuals from the
general adult population and face-to-face interviews.
However, the surveys differ in terms of their research topics and methodological
details. (More information is given in the appendix.) The majority of these
surveys were conducted as part of the ISSP and were organised by the Institute
of Sociology of the Academy of Sciences of the Czech Republic, so there are a
number of similarities in the methods used and in the background of the sur-
veys.5
In the case of Czechoslovak surveys, only the data for the Czech part of the
survey are presented.
Research surveys carried out at the start of the 1990s exhibited in comparative
terms a relatively low level of non-response. The Social Stratification in Eastern
Europe (SSEE) study of 1993, which had the highest rate of non-response
4
The World Association for Public Opinion Research (WAPOR), the World Association
for Marketing Research (ESOMAR), the European Federation of Associations of Market
Research Organisations (EFAMRO), and the Association of Market and Opinion Research
Agencies (SIMAR). The main Czech research agencies are members of at least one of these
organisations.
5
The International Social Survey Programme is a long-term comparative research project,
in which each year a survey with an identical module of questions is conducted in each of
the participant countries. As of 2006 there are 41 member countries. The Czech Republic
has been conducting the ISSP survey since 1992; all Czech ISSP surveys are based on probability
samples. See the ISSP website: http://www.issp.org.
socrev2007-1.indb 569socrev2007-1.indb 569 3.9.2007 18:41:313.9.2007 18:41:31
Sociologický časopis/Czech Sociological Review, 2007, Vol. 43, No. 3
570
among surveys conducted in the first half of the 1990s, was accompanied by a
negative media campaign. As a result an initially low non-response rate increased
dramatically to a final overall average of 35% [Matějů 1993]. Non-response rates
in other contemporary surveys fluctuated between 17% and 30%. This outcome
is comparable to the non-response rates recorded in the General Social Survey
(GSS) in the United States, which in the years between 1975 and 2004 ranged
from 17% to 30% [Davis, Smith and Marsden 2005]. However, with the exception
of the SSEE, the outcomes in this period were significantly better than the German
Allbus study, which had a non-response rate of 32­48% between 1988 and
1995 [Allbus Datenservice 2007]. It should be noted that the GSS and Allbus are
among the most important social-science surveys in their respective countries
and both put emphasis on the use of a rigorous methodology.
Although some of the differences in response rates can be ascribed to methodological
differences and differing survey topics, Figure 1 clearly shows that
there was a substantial increase in the non-response rate in the mid-1990s. In
surveys conducted after 1995 the rate of non-response did not fall below 30%.
Regrettably it was not unusual for non-response rates to exceed 50%. The data
presented in Figure 1 confirm the view expressed within the Czech survey reFigure
1. Non-response in selected academic surveys in the Czech Republic,
1991­2005 (%)
Note: For additional information on surveys see the appendix. Non-response does not
include ineligible addresses, i.e. out-of-sample and vacant housing units and non-residential
addresses.
Source: Sociological Data Archive, Institute of Sociology AS CR.
3.71
6.12
0.53
4.82
0.81
0.03
0.22
4.25
1.34
5.74
4.06
4.44
9.33
4.24
6.64
1.35
9.15
4.24
0
01
02
03
04
05
06
07
PJSI
1
99
1
PSSI
1
99
2SSV
E
91
93
PSSI
1
99
3
PSSI
1
99
4
PSSI
1
99
5
PJSI
1
99
5
PSSI
1
99
6
PSSI
1
99
7
PSSI
1
99
9
PSSI
1
99
8
PSSI
2
00
0
PSSI
2
00
1
PSSI
2
00
2
PSSI
2
00
3
PSSI
2
00
4
PSSI
2
00
5
3.71
%
6.12
0.53
4.82
0.81
0.03
0.22
4.25
1.34
5.74
4.06
4.44
9.33
4.24
6.64
1.35
9.15
4.24
socrev2007-1.indb 570socrev2007-1.indb 570 3.9.2007 18:41:313.9.2007 18:41:31
Jindřich Krejčí: Non-Response in Probability Sample Surveys in the Czech Republic
571
search community that in the 1990s there was indeed a sharp decline in response
rates. In fact, our data reveal that after 1995 response rates never exceeded 70%.
Equally important is the fact that the belief that there had been an increase in the
non-response rate after 1996 is not supported by the evidence examined.
In Figure 2, which provides a more detailed view of the ISSP surveys conducted
between 1997 and 2005, we can see that the differences in response rates
are not just caused by changes in the number of refusals and thus the willingness
of the public to cooperate, but also stem from differences in the number of noncontacts
and other forms of non-response. It is interesting to note that non-contacts
do not exhibit a consistent trend, and it would therefore be dubious to try to
explain their variations as resulting from external causes alone.
The figures for other forms of non-response are also relatively high. For
example, in the ISSP 1999, which had the highest non-response rate, the main
source of the problem was failures on the part of the interviewers [STEM 1999].
We can therefore infer that the reason for the considerable differences in response
rates, rather than being evidence of a negative trend in the Czech public's willingness
to cooperate, are the result of other factors, such as the performance of the
Figure 2. Refusals, non-contacted and other unproductive interviews in Czech ISSP
surveys, 1997­2005 (%)
Note: For additional information on ISSP surveys see the appendix. ISSP (1998) is not
included owing to missing information on non-contacts.
Source: Sociological Data Archive, Institute of Sociology AS CR.
%001%09%08%07%06%05%04%03%02%01%0
5002PSSI
4002PSSI
3002PSSI
2002PSSI
1002PSSI
0002PSSI
9991PSSI
7991PSSI
lasufer non-contacted evitcudorpnurehto evitcudorp
3.81
5.42
2.62
8.52
9.12
0.22
4.42
0.517.02
4.5
3.91
3.2
5.3
4.4
9.5
3.3
9.3
8.11
0.52
4.32
4.61
6.41
6.8
1.61
socrev2007-1.indb 571socrev2007-1.indb 571 3.9.2007 18:41:313.9.2007 18:41:31
Sociologický časopis/Czech Sociological Review, 2007, Vol. 43, No. 3
572
fieldwork agency, the methodology used, timing, and other conditions affecting
data collection.
An important factor is certainly the survey topic. For example, the technical
report for the ISSP 1996 on `The Role of the Government' explains the low
response rate as a result of the unwillingness of respondents to answer questions
on political topics in a period following two national elections [STEM 1996]. Also,
the response rates for the ISSP 1998 on `Religion' were affected by the timing of
data collection ­ during the summer holiday period ­ and by the topic of the research
itself. This is evident from biases in the socio-demographic profile of the
sample. There was an over-representation of respondents claiming to belong to a
specific religious denomination, a higher than expected level of religiosity, and a
relatively high number of Christian Democrat (KDU-ČSL) voters.
It is appropriate at this point to briefly examine an official statistical survey.
Here we will look at the Czech Statistical Office's quarterly `Labour Force
Survey' (LFS ­ Výběrové šetření pracovních sil) which is a large household survey
(of approximately twenty-six thousand, i.e. 0.6% of dwelling households in the
Czech Republic) that examines the position of respondents in the labour market,
labour mobility, economic activities in the population, social status, general economic
conditions, and other related topics. The results are used for the purposes
of official statistics and are available on the website and in the publications of
Figure 3. Non-response in the quarterly Labour Force Survey in the Czech Republic,
2002 q1 to 2005 q4 (%)
Source: Labour Force Survey, Czech Statistical Office.
0
5
01
51
02
52
03
53
4Q3Q2Q1Q4Q3Q2Q1Q4Q3Q2Q1Q4Q3Q2Q1Q
30022002 4002 5002
total non-response
refusals
non-contacts
other non-response
%
8.42
6.42
8.621.629.62
1.82
1.03
8.62
1.22
1.22
0.12
9.02
4.12
7.228.12
3.22
socrev2007-1.indb 572socrev2007-1.indb 572 3.9.2007 18:41:313.9.2007 18:41:31
Jindřich Krejčí: Non-Response in Probability Sample Surveys in the Czech Republic
573
the Czech Statistical Office and Eurostat.6
The methodology used corresponds to
the requirements of the ILO and Eurostat and is based on a random sampling of
households7
and face-to-face interviews.
Figure 3 presents data from the LFS for the period since the last major revision
in survey methodology in 2002. Unfortunately detailed information on
outcome rates are available only since 2002; between 1992 and 2001 only total
non-response rates were reported and therefore this period cannot be included
in the figure.
One striking feature of this data is that non-response rates in the LFS are
considerably lower than in the academic surveys shown in Figure 1. The LFS
non-response rates fluctuate between 20% and 30%. This evidence demonstrates
that in the current survey climate in the Czech Republic it is possible to attain
relatively high survey response rates.
Of course, in this case, we can also suppose some impact from the `sponsorship
effect', that is, that the willingness to cooperate depends on the data collection
agency and the purpose of its activities. Consequently, official statistics
often have lower non-response than academic surveys, and academic surveys
commonly have lower non-response than commercial polls [Groves and Couper
1998]. On the other hand, the differences in outcomes do not just stem from refusals,
which are only an effect of the willingness factor. Figure 3 also clearly reveals
a general trend toward decreasing non-response rates, both in the case of refusals
and non-contacts. We can therefore conclude that the data at our disposal do not
support the prevailing conviction that all survey response rates are declining in
the Czech Republic.
The methodology used in the ESS I and ESS II in the Czech Republic
and their response rates
ESS is a biennial academic study of changing social, political, and economic attitudes
within Europe. A total of twenty-six countries took part in the first two
waves of research. The project has an elaborate organisational structure that is
partly funded by the European Commission and the European Science Foundation.
However, the surveys conducted in the individual participant countries are
sponsored from national sources. Detailed information on the project, along with
data and documentation for particular national surveys, are available on the official
ESS website.8
6
Czech Statistical Office: http://www.czso.cz; Eurostat: http://ec.europa.eu/eurostat.
7
The Czech Statistical Office manages its own database of households, which is organised
as the outcome of a census and is used for sample surveys as a sampling frame. Before
2001 this database was used also in a number of academic surveys. The major problem is
connected to the recency of the database, as the number of wrong addresses rises with the
time passed since the census. The last Czech census was organised in 2001.
8
http://www.europeansocialsurvey.org.
socrev2007-1.indb 573socrev2007-1.indb 573 3.9.2007 18:41:323.9.2007 18:41:32
Sociologický časopis/Czech Sociological Review, 2007, Vol. 43, No. 3
574
The European Social Survey (ESS)
The key objective of the ESS is to build up a database of time series and internationally
comparable indicators that can be easily accessed by all researchers,
including project coordinators, under the same conditions. The ESS is therefore
one of the most important infrastructures in European social research. It uses a
general model, which is applied repeatedly in all research waves, and it contains
a large methodological component and specific thematically defined modules.
The ESS strives toward standardisation and toward reducing the differences
in research design and fieldwork procedures in participant countries, and it demands
adherence to the specific survey parameters [ESS 2002, 2004]. These parameters
include, for example, full coverage of the residential population aged 15
and over, use of strict probabilistic methods of sampling at all stages, a minimum
effective sample size of one thousand five hundred respondents, a target response
rate of 70%, and a target non-contact rate of 3%. The Central Coordination Team
must approve the surveying methodology proposed by national survey teams.
In order to get an idea of the rigorous surveying protocols the ESS insists on,
let us examine for a moment their recommendations for achieving high response
rates [Philippens and Billet 2004; Philippens et al. 2003; Pleyser and Billet 2006].
* The fieldwork period should last at least one month
* There should be briefing and training of all interviewers
* Interviewer workloads should be limited (i.e. a maximum of 24 interviews per
interviewer)
* Face-to-face data collection should be used
* At least four calls should be made on different days and at different times
when trying to interview difficult to reach potential respondents
* In countries with no individual level sampling frames all first contacts for
interviews should be made face-to-face
* Attempted interviews should be spread over at least two different weeks
* No substitution of units selected for interview at any stage
* Refusal conversion strategies should be used
* Detailed contact forms should be used
* Detailed monitoring of the data collection process must be provided
* Quality control checks of interviews completed should be adopted
However, it should be noted that national ESS survey coordinators have the
option of proposing strategies for reducing non-response rates. In this respect,
the ESS recommends using advance letters of introduction, toll-free telephone
numbers for potential respondents to contact, selecting experienced interviewers,
extra training of interviewers in response-maximisation techniques and doorstep
interactions, implementing refusal avoidance and conversion techniques, re-consocrev2007-1.indb
574socrev2007-1.indb 574 3.9.2007 18:41:323.9.2007 18:41:32
Jindřich Krejčí: Non-Response in Probability Sample Surveys in the Czech Republic
575
tacting `soft' refusals and `indecisive' non-contacts, and the use of appropriate
respondent incentives and other strategies.
The European Social Survey in the Czech Republic
Two waves of the ESS have been conducted so far in the Czech Republic: the ESS I
in 2002­2003, and the ESS II in 2004. Like the ISSP, both surveys were organised
through the Institute of Sociology of the Academy of Sciences of the Czech Republic.
Most of the main methodological features prescribed by the ESS were observed.
It is important at this point to outline five key areas of difference between
the successive waves of the ESS.
a) Fieldwork agencies
Data collection in the first wave of research was conducted by STEM and in the
second wave by SC&C. Both companies are among the leading research agencies
in the Czech Republic. Both companies have long-established, extensive interviewer
networks. Moreover, both have conducted numerous probability surveys
and have participated in a number of academic research projects, including highlevel
international research programmes.9
b) Sampling method
Probability sampling or more specifically stratified multistage sampling was
used in both waves, but the specific methods applied differed. In the ESS I, the
sampling frame used by STEM was derived from the SIPO database.10
The stratification
factors were communities and region. In the first step, the communities
were selected, then the SIPO addresses (i.e. households), and in the third step
individual respondents were selected with the aid of a Kish grid.
9
For example, STEM undertook the ISJP (1991, 1995) and the ISSP (1992, 1996, 1997, 1999),
while SC&C was responsible for the ISSP (1998, 2000­2004), SIALS (1998), and the EVS
(1999).
10
The SIPO is a special payment system that in communist Czechoslovakia provided
services that are in other countries often provided by banks, and it is still widely used
today. Regular household bill payments, especially utility payments, TV and radio fees,
telephone bills, and bills for other services, are grouped into a single bill and are paid
through the post office into the relevant accounts. The continuous updating of registered
addresses using the SIPO system initially represented an almost comprehensive resource
for household sampling with an estimated coverage of up to 98% of Czech households (in
the first half of the 1990s). There now exist other means and systems of payments, and the
number of addresses in the SIPO is much lower than the number of households recorded
in official statistics, and therefore the quality of this database source is questionable.
socrev2007-1.indb 575socrev2007-1.indb 575 3.9.2007 18:41:323.9.2007 18:41:32
Sociologický časopis/Czech Sociological Review, 2007, Vol. 43, No. 3
576
In the ESS II the sampling frame used was the UIR-ADR address register,
i.e. an electronic database of all buildings in the Czech Republic.11
The register is
regularly updated and identifies buildings on individual streets (where streets
can be identified) within all towns and villages. Sampling proceeded in four
stages: 1) communities, 2) streets or small communities, 3) household units, and
4) individuals. The unique character of this sampling frame produced a complicated
research design. For example, the database did not provide information on
the number of household units per street, only the number of buildings, so in the
third step of sampling a list of household units in selected streets and small communities
had to be drawn up first, following a preliminary interviewer appraisal
of the target street.
c) Financing
The method of financing in each round of the survey differed. In the first wave,
the Ministry of Education, Youth and Sport provided a grant for the research
under the International Non-Governmental Organisation (INGO) programme.
In the second wave, the Czech Science Foundation provided funding. In the first
wave the funds were only released at the end of 2002 and less was provided than
had been planned for in the approved grant. This had a serious impact on the
preparation and the course of data collection, the performance of which had been
sub-contracted. As a result, some of the techniques that had been planned could
not be used owing to time and financial constraints. In addition, data collection
dragged on into 2003, and cooperation with the fieldwork agency also proved to
be complicated.
d) Fieldwork timing
For the financial reasons mentioned earlier, data collection in the ESS I was conducted,
contrary to the instructions of the research coordinating team, in two stages:
from 21 November to the end of December 2002, and from 2 January to 9 March
2003. In the ESS II, data collection was conducted within the required time frame
and took place in a single period, between 1 October and 13 December 2004.
e) Methods for enhancing the response rate
In the ESS I, owing to financial and time constraints, it was not possible to carry
out the special training for interviewers that had been planned or to apply specific
strategies designed to enhance response rates. It was therefore necessary to
rely on the standard training provided to interviewers by STEM. The interviewers
visited the sampled addresses personally and advance letters of introduction
11
The Register, organised by the Ministry of Labour and Social Affairs, is publicly accessible
on the Web.
socrev2007-1.indb 576socrev2007-1.indb 576 3.9.2007 18:41:323.9.2007 18:41:32
Jindřich Krejčí: Non-Response in Probability Sample Surveys in the Czech Republic
577
were used. Special procedures for refusal conversions were not applied beyond
the regular methods used by STEM interviewers. Also, no special incentives were
used. Remuneration for interviewers paid for the return of completed questionnaires
was below the average for the tasks they performed. The methodology
proposed a minimum of six attempts at contacting respondents on various times
and days in the week.
However, some of these procedures were not followed. For example, after
data collection was completed it was discovered that fewer than four attempts
had been made at 46% of the non-contact addresses. A large number of contact
forms were completed incorrectly, and it was difficult to reconstruct how the
fieldwork had proceeded [see also Phillipens et al. 2003].
In the ESS II, approximately three hours of special training was provided to
interviewers. Emphasis was placed on ensuring that the contact forms would be
filled in properly. Only experienced interviewers from SC&C who have regular
training in techniques for establishing contacts with respondents and maximising
refusal conversions carried out the interviews. Advance letters of introduction
were used, and the households contacted were provided with the agency's
telephone number. Payments to interviewers for completed and incomplete questionnaires
differed significantly. They received higher than usual remuneration
for their work and were rewarded with a bonus for achieving high response rates
(70%). At the same time emphasis was put on monitoring the interviewers' work.
More than 30% of the interviews were supervised and 572 sample units were
back-checked (except for 15 cases in which back-checks were not achieved the
outcomes from the fieldwork were confirmed). To elicit good will the respondents
were given a large package of coffee or tea and a calendar. In addition, at last
four attempts were made to contact respondents. Data collection was controlled
on an ongoing basis and `soft' refusals and some non-contacts were targeted for
additional `conversion' visits by interviewers in order to reduce non-response
rates to below 30%.
f) Evaluation
In formal terms the ESS I used a simpler and cleaner sampling method than the
ESS II. The procedure used to collect data was organisationally less demanding
and there was hence less room for error. Nevertheless, as noted above, the sampling
frame derived from the SIPO database is problematic given the probably
limited population coverage. There are therefore problems in assessing the impact
of sampling design effects on total survey error for the ESS I. In contrast to
other Czech academic surveys, such as the ISSP, the ESS I was organised without
any specific techniques to enhance the response rate. Unsurprisingly, this led to
problems with the interviewer team, many of whom had difficulties completing
the contact forms correctly.
In the ESS II no satisfactory sampling frame was available. Consequently, a
complicated sampling procedure was used that placed high demands on organissocrev2007-1.indb
577socrev2007-1.indb 577 3.9.2007 18:41:323.9.2007 18:41:32
Sociologický časopis/Czech Sociological Review, 2007, Vol. 43, No. 3
578
ing the research. For example, as noted above, the breakdown of dwelling units
on selected streets had to be obtained before interviewing commenced. However,
there was one key advantage to performing the somewhat onerous task of creating
a `fresh' sampling frame ­ the interviewers' efforts to contact respondents
were not complicated by the problem of incorrect addresses. Moreover, the interviewers
underwent additional training specifically for the ESS II project. According
to the SC&C manager in charge of the project the higher response rate
achieved in the fieldwork stemmed from a combination of three main factors:
i) the use of the appropriate respondent incentives, ii) the higher rewards offered
to interviewers, and iii) the elaborate system implemented to explain the purpose
of the survey to the respondents.
At this point it should be stressed again that effort and particular methods
for enhancing response rates can also have a negative influence on survey biases.
For example, the positive and negative effects of respondent incentives on total
survey errors are largely examined and discussed in analytical literature,12
giving
bonuses to interviewers for high response rates can encourage them to falsify
interviews, conversion methods can have problematic impact on quality of data,
etc. Therefore only well-considered strategies can reduce risks and have a positive
effect on total survey error. Finally, it should also be mentioned that in many
cases the relationship between stronger efforts to enhance the response rate and a
reduction of total bias was confirmed [e.g. Olson 2006; Lynn et al. 2002].
The effects on response rates
A comparison of the outcome rates in both waves of the Czech ESS surveys is
presented in Table 1. Row O of this table shows the data on total response rates
calculated in accordance with current AAPOR standards [2006]. It is immediately
obvious that the response rate for the ESS II was 12.2 percentage points higher
than the response rate for the first wave of the ESS. This is a substantial difference,
which stems from the fact that on most criteria presented in Table 2 the ESS II
was more effective, e.g. had a lower rate of refusals. While the effectiveness of the
ESS II in reducing non-response is dramatic, the response rate of 55.5% is similar
to the achievements of other academic surveys undertaken in recent years (e.g.
ISSP 2000­2003, see Figure 1.) where no specific strategy for attaining a higher
response rate was used.
The picture changes when we look at the composition of non-responses. In
the ESS II questionnaires were not distributed at all to almost one thousand two
12
A review of this literature can be found, for example, in Stoop [2005], who summarises
that `the evidence from many studies...is not entirely conclusive', but `most researchers
seem to agree that a small incentive is generally seen as a nice gesture, both by interviewers
and respondents, that a large incentive is not necessarily more effective than a small
one, and that incentives do not have a detrimental effect on survey quality' [ibid.: 94].
socrev2007-1.indb 578socrev2007-1.indb 578 3.9.2007 18:41:323.9.2007 18:41:32
Jindřich Krejčí: Non-Response in Probability Sample Surveys in the Czech Republic
579
hundred of the selected addresses (row D). If we take this fact into account, the
response rate achieved during fieldwork, as the estimate shown in row N reveals,
was 71%.
Based on experience from the ESS I, preparations for the second wave anticipated
a low response rate and the size of the initial ESS II sample was designed to
accommodate this expectation. The better than expected response rate achieved
in the second wave meant that some of the addresses could not be used. This was
because the budget for the ESS II would not have covered the expense involved in
conducting the additional interviews and processing the resulting data. In fact, in
the ESS II the largest proportion of the recorded non-response (as shown in rows
D and F-J in Table 1) stemmed from the underestimation of the actual response
rate in the preparation of the survey. This underestimation error accounted for
more than 21% of the survey sample and was paradoxically a product of the success
of rather than any failure in the fieldwork.
The standard definitions of the indicators of response and non-response
rates [AAPOR 2006; Groves 2004: 135­145] do not take this kind of unusual erTable
1. Response and non-response rates in ESS surveys in the Czech Republic
ESS I 2002/2003 ESS II 2004
N % N %
A. Total issued sample 3330 100.0 5531 100.0
B. Wrong addresses 191 5.7 66 1.2
C. Respondent deceased/moved 0 0.0 8 0.1
D. Sampled addresses not used 0 0.0 1196 21.6
E.
Eligible addresses used for fieldwork:
A-( B+C+D)
3139 94.3 4261 77.0
F. Respondent unable to co-operate 25 0.8 14 0.3
G. Respondent unavailable 142 4.3 33 0.6
H. Refusals 628 18.9 601 10.9
I. No contacts 365 11.0 587 10.6
J. Other non responses 619 18.6 0 0.0
K. No. of achieved interviews 1360 40.8 3026 54.7
L. Co-operation rate: K/(H+K) 68.4 83.4
M. Contact rate: K/(G+I+K) 72.8 83.0
N. Field response rate: K/E 43.3 71.0
O. Response rate: K/[A-( B+C)] 43.3 55.5
Source: ESS I, ESS II (Institute of Sociology AS CR, ESS).
socrev2007-1.indb 579socrev2007-1.indb 579 3.9.2007 18:41:323.9.2007 18:41:32
Sociologický časopis/Czech Sociological Review, 2007, Vol. 43, No. 3
580
ror into account. In the case of multistage stratified sampling it is a questionable
practice to eliminate any selected units after the sample has been drawn, especially
during the interviewing stage, even if we assume that the decision of which
addresses to use and which to eliminate was made randomly. That is why in the
overall evaluation of the quality of the data from the ESS II the Czech Republic is
not ranked among the countries with a response rate of at least 70%. The unused
addresses are included under non-contacts in the calculations for this indicator
[see Pleyser and Billiet 2006].
However, it is also misleading to evaluate the contact rate including the
unused addresses, when no attempt was ever made at contacting respondents at
these addresses. Therefore, a deeper analysis of non-response and conclusions on
the ability to contact respondents and the willingness of the public to cooperate
should be based on the responses the survey actually achieved in the field. In the
following analysis we will focus mainly on outcomes achieved during the fieldwork.
Pleyser and Billiet [2006] proceeded similarly in their analysis of the quality
of the ESS international data file and did not include the unused addresses in
the evaluation of non-contacts and refusals.
As Figure 1 shows, the field response rate of 71% in the ESS II greatly exceeded
the response rates attained in contemporary Czech academic surveys and
Figure 4. Cooperation and contact rates in ESS and ISSP surveys (1997­2005)
in the Czech Republic (%)
Note: For additional information on ISSP surveys, see the appendix. The ISSP (1998) is
not included owing to missing information on non-contacts.
Source: Sociological Data Archive, ESS I, ESS II (Institute of Sociology AS CR, ESS).
0010908070605040302010
5002PSSI
4002PSSI
3002PSSI
2002PSSI
1002PSSI
0002PSSI
9991PSSI
7991PSSI
IISSE
ISSE
cooperation rate
contact rate
socrev2007-1.indb 580socrev2007-1.indb 580 3.9.2007 18:41:323.9.2007 18:41:32
Jindřich Krejčí: Non-Response in Probability Sample Surveys in the Czech Republic
581
is closer to the results achieved in the Labour Force Survey (shown in Figure 3).
Equally importantly, this evidence demonstrates that the survey climate is not as
unfavourable as it is feared to be. The data presented in row L of Table 1 show
that the cooperation rate in the ESS II was 83.4%. Compared to the ESS I this
represents an increase of almost 15 percentage points. Moreover, if we examine
Figure 4 we see that this rate is significantly higher than the cooperation rates for
all ISSP surveys since 1997.
The evidence presented in row N of Table 1 reveals that the ESS II had a
higher contact rate than its predecessor. Moreover, Figure 4 reveals that the contact
rate in the ESS II is also relatively high when compared to the results of other
Czech academic surveys.
It is interesting to look at the number of non-contacts in relation to other
forms of non-response and in relation to the interviews conducted. The non-contact
rate was 13.6% in the ESS II, which is the highest percentage of non-contacts
among the countries that took part in the survey. Pleyser and Billiet [2006: 15]
ascribe this to the complicated sampling procedure. If we compare the ESS II with
recent ISSP surveys, which are characterised by relatively simple sample designs,
we see that Czech surveys often have high non-contact rates. Such comparative
evidence confirms that non-contact troubles during interviewing are part of a
more general problem. However, not being able to contact potential respondents
is not an insurmountable problem. This is evident from the quarterly Labour
Force Surveys conducted by the Czech Statistical Office. The data presented in
Figure 3 show that the non-contact rate in the LFS since 2002 has generally been
below 10%.
If we look at row J of Table 2 for the ESS II we see that it is also possible to
completely eliminate other forms of non-response in the course of the fieldwork.
According to a post-ESS I survey report written by STEM, the high number of
`other non-responses' in this survey consisted of various administrative errors
made in the organisation of data collection and errors in data processing. One
key problem was questionnaires not being returned by interviewers. In surveys
such as the Czech Labour Force Survey, the GSS and Allbus, where data quality
is an explicit goal, errors of this type rarely occur.
An international comparison also reveals differences in each wave of the
Czech ESS survey. With regard to the non-response rate, the Czech survey in
2002­2003 had the third worst outcome of all countries surveyed. The Czech survey's
total response rate in 2004 also ranked in the lower half of all the countries
surveyed. Nonetheless, it was better than that of countries such as the UK,
Germany, and Spain. However, if we examine field response rates, which better
reflect the actual level of cooperation of the population, the Czech Republic was
among the countries with the best results ­ exceeding the target level of 70%. Notwithstanding
the failure to accurately estimate the sample size at the outset, the
strategy applied in the ESS II to enhance response rates was effective. The field
response rate was increased beyond the customary level achieved in academic
socrev2007-1.indb 581socrev2007-1.indb 581 3.9.2007 18:41:333.9.2007 18:41:33
Sociologický časopis/Czech Sociological Review, 2007, Vol. 43, No. 3
582
surveys in the Czech Republic and well above the international average attained
in the ESS. This outcome resulted from a combination of improvements in the
monitored sources of non-response ­ refusals, non-contacts, and other miscellaneous
forms of non-response.
Conclusion
The above assessment of Czech probability surveys conducted for academic purposes
since 1995 shows that response rates are a serious problem. This could be
reason to question the reliability of the data on Czech society and the analyses
based on these data. The differences between the response rates during the first
half of the 1990s and in surveys conducted later on point to a deterioration of
the survey climate during this period. There have been consistently high rates
of refusals and non-contacts. The level of non-contacts recorded in the Czech
Republic rank among the highest measured internationally. However, the conventional
opinion that there is a declining secular trend in response rates since
the mid-1990s was not confirmed by available survey evidence.
Our examination of academic surveys demonstrates that a large part of the
non-response problem stems from research imperfections. Outcome rates often
include a big portion of the category `other unproductive', that is, administrative
and organisational failures. Interestingly, the trend in non-contact rates fluctuates
considerably. We may therefore conclude that, in addition to the unfavourable
survey climate, low response rates are largely the result of factors connected with
the performance of the fieldwork agency. Fortunately, these issues can often be
addressed by upgrading fieldwork protocols and methodologies. We also found
that it is possible to obtain substantially better outcomes in survey research. This
is particularly evident in the Labour Force Survey conducted by the Czech Statistical
Office and in the second wave of the European Social Survey.
Our comparison of the two waves of the ESS did not look for differences
in data quality between both datasets and did not provide an evaluation of the
effects of particular survey strategies on total biases. There are numerous factors
connected with both waves of the ESS that were capable of producing errors.
The appropriate analysis of the impact of response rate enhancement on total
survey error would therefore necessarily involve not only simple comparisons
of the composition of surveyed samples but also an examination of the differences
between respondents and non-respondents. Unfortunately, the limited size
of this article puts such an analysis beyond its scope, but the article does generate
a starting point for future study. Therefore, we cannot conclude here whether the
described strategy of response rate enhancement was successful in terms of survey
accuracy. Nonetheless, based on the evidence provided we can point out the
need to deal with the non-response problem and argue for the use of sensible and
prudent survey strategies that take all types of survey error into consideration.
However, such strategies often require greater effort and more investment.
socrev2007-1.indb 582socrev2007-1.indb 582 3.9.2007 18:41:333.9.2007 18:41:33
Jindřich Krejčí: Non-Response in Probability Sample Surveys in the Czech Republic
583
In the academic sphere traditional approaches to organising survey research
have focused on sampling and related errors. But contrary to expectations,
sampling design effects are not the only source of survey error. Total survey error
is a complex problem and in order for us to obtain more accurate results it is
necessary to focus on all of its components: sampling error, coverage error, errors
of measurement and non-response error. Moreover, besides being accurate data
must also be timely, coherent, accessible, comparable, and fit for use. Finally, as
Groves [2004] has shown, the quality of a sample survey derives from its efficiency,
that is, from the relationship between accuracy and the costs of achieving it.
For all these reasons recent concepts of the quality of survey research have moved
away from purely statistical approaches to accuracy toward multidimensional
approaches that involve quality management of the survey process [see Biemer
and Lyberg 2003].
In the Czech Republic survey non-response can be a serious obstacle to reliability.
Consequently, when organising surveys it is prudent to devote some
thought to how much effort should be invested in attempts to reduce it. Ignoring
this problem devalues the resources expended on reducing other types of error.
Successful survey research strategies require more than just formal and selective
approaches to data quality enhancement. This issue is not merely a topic of importance
for fieldwork agencies, but is also one of key concern for all coordinators
of research projects, because it is they who formulate the methodological requirements
for many probability sample surveys.
JINDŘICH KREJČÍ is the head of the Sociological Data Archive (SDA) at the Institute of
Sociology. His research focuses on survey methodology, data quality, and the study of
political attitudes. He is the editor of the book Kvalita výzkumů volebních preferenci
(The Quality of Survey Research on Electoral Preferences) and the author of several articles
in scientific journals and books. At the Faculty of Social Sciences of Charles University
he teaches courses in survey methodology, methods of social data analyses, and data
management. He has also participated in the surveys of the International Social Survey
Programme (ISSP) and the European Social Survey (ESS).
References
AAPOR. 2006. `Standard Definitions. Final Dispositions of Case Codes and Outcome
Rates for Surveys.' Lenexa, KS: The American Association for Public Opinion
Research. Retrieved 25 May 2007. (http://www.aapor.org/pdfs/standarddefs_4.pdf).
Allbus Datenservice. 2007. Allbus Web presentation, codebook and data download at the
GESIS Zentralarchiv für Empirische Sozialforschung, Cologne. Retrieved 31 May 2007
(http://www.gesis.org/Datenservice/ALLBUS).
Baim, Julian. 1991. `Response Rates: A Multinational Perspective.' Marketing and Research
Today 19 (2): 114­120.
socrev2007-1.indb 583socrev2007-1.indb 583 3.9.2007 18:41:333.9.2007 18:41:33
Sociologický časopis/Czech Sociological Review, 2007, Vol. 43, No. 3
584
Biemer, Paul P. and Lars E. Lyberg. 2003. Introduction to Survey Quality. Hoboken: John
Wiley & Sons.
Bogart, Leo. (ed.) 1987. `The Future Study of Public Opinion: A Symposium.' Public
Opinion Quarterly 51 (2: Supplement): S173­S191.
Bradburn, Norman M. 1992. `Presidential Address: A Response to the Nonresponse
Problem.' Public Opinion Quarterly 56: 391­397.
Davis, James A., Tom W. Smith and Peter V. Marsden. 2005. `General Social Surveys,
1972­2004. Codebook Appendices.' Ann Arbor: ICPSR. Retrieved 20 October 2006
(http://www.icpsr.umich.edu; Cat. No. 6293070).
de Heer, Wim 1999. `International Response Trends: Results of an International Survey.'
Journal of Official Statistics 15 (2): 129­142.
de Leeuw, Edith and Wim de Heer. 2002. `Trends in Household Survey Nonresponse:
A Longitudinal and International Comparison.' Pp. 41­54 in Survey Nonresponse,
edited by Robert M. Groves, Don A. Dillman, John L. Eltinge and Roderick J.A. Little.
New York: John Wiley & Sons.
Dillman, Don A., John L. Eltinge, Robert M. Groves and Roderick J.A. Little. 2002.
`Survey Nonresponse in Design, Data Collection, and Analysis.' Pp. 3­26 in Survey
Nonresponse, edited by Robert M. Groves, Don A. Dillman, John L. Eltinge, and
Roderick J.A. Little. New York: John Wiley & Sons.
ESS. 2002. European Social Survey, Round 1: Specification for Participating Countries.
European Social Survey Web. London: Centre for Comparative Social Surveys.
Retrieved 16 March 2006 (http://www.europeansocialsurvey.org).
ESS. 2004. European Social Survey, Round 2: Specification for Participating Countries.
European Social Survey Web. London: Centre for Comparative Social Surveys.
Retrieved 16 March 2006 (http://www.europeansocialsurvey.org).
Groves, Robert M. 2004 (first edition 1989). Survey Errors and Survey Costs. Second
edition. Hoboken: John Wiley & Sons.
Groves, Robert M. 2006. `Nonresponse Rates and Nonresponse Bias in Household
Surveys.' Public Opinion Quarterly 70 (5): 646­675.
Groves, Robert M. and M. Couper. 1998. Nonresponse in Household Interview Surveys.
New York: John Wiley & Sons.
Joye, Dominique. 2004. `Pravděpodobnostní výběry a pravděpodobnost návratnosti.'
(Probability Samples and Probability of Response) Pp. 185­201 in Kvalita výzkumů
volebních preferencí, edited by J. Krejčí. Prague: Sociologický ústav AV ČR.
Kaase, Max. (ed.) 1999. Qualitätskrieterien der Umfrageforschung ­ Quality Criteria for
Survey Research. Berlin: Akademie Verlag GmbH.
Keeter, Scott, Carolyn Miller, Andrew Kohut, Robert M. Groves and Stanley Presser.
2000. `Consequences of Reducing Nonresponse in a National Telephone Survey.'
Public Opinion Quarterly 64 (2): 125­148.
Krejčí, Jindřich. 2004. `Podmínky pro realizaci výběrových šetření v České republice.'
(Conditions for Conducting Sample Surveys in the Czech Republic) Pp. 67­86 in
Kvalita výzkumů volebních preferencí, edited by J. Krejčí. Prague: Sociologický ústav
AV ČR.
Krejčí, Jindřich, Oľga Gyarfášová, Dominique Joye, František Kalvas, Martin Kreidl,
Vladimír Krivý, Daniel Kunštát, Tomáš Lebeda, Yana Leontiyeva and Klára
Plecitá-Vlachová. 2004. Kvalita výzkumů volebních preferencí. (The Quality of Surveys
on Electoral Preferences) Prague: Sociologický ústav AV ČR.
Lynn, Peter, Paul Clarke, Jean Martin and Patric Sturgis. 2002. `The Effect of Extended
Interviewer Efforts on Nonresponse Bias.' Pp. 149­162 in Survey Nonresponse, edited
by Robert M. Groves, Don A. Dillman, John L. Eltinge and Roderick J.A. Little. New
York: John Wiley & Sons.
socrev2007-1.indb 584socrev2007-1.indb 584 3.9.2007 18:41:333.9.2007 18:41:33
Jindřich Krejčí: Non-Response in Probability Sample Surveys in the Czech Republic
585
Matějů, Petr. 1993. Social Stratification in Eastern Europe after 1989: Report on Data Collection
for the General Population Survey in the Czech Republic. Technical Report from the Survey.
Prague: Sociologický ústav AV ČR.
Merkle, Daniel and Murray Edelman. 2002. `Nonresponse in Exit Polls: A Comprehensive
Analysis.' Pp. 243­258 in Survey Nonresponse, edited by Robert M. Groves, Don
A. Dillman, John L. Eltinge and Roderick J.A. Little. New York: John Wiley & Sons.
Olson, Kristen. 2006. `Survey Participation, Nonresponse Bias, Measurement Error Bias,
and Total Bias.' Public Opinion Quarterly 70 (5): 737­758.
Philippens, Michel and Jaak Billiet. 2004. `Monitoring and Evaluating Non-response
Issues and Fieldwork Efforts in the European Social Survey.' Paper presented
at the European Conference on Quality and Methodology in Official Statistics,
Mainz, Germany, 24­26 May 2004. European Social Survey Web. London:
Centre for Comparative Social Surveys. Retrieved 16 March 2006 (http://www.
europeansocialsurvey.org).
Philippens, Michel, Geert Loosveldt, Ineke Stoop and Jaak Billiet. 2003. `Noncontact
Rates and Interviewer Calling Strategies in the ESS.' Paper presented at the
International Workshop on Household Nonresponse, Leuven, Belgium, 22­24
September 2004. European Social Survey Web. London: Centre for Comparative
Social Surveys. Retrieved 16 March 2006 (http://www.europeansocialsurvey.org).
Pleyser, Stefaan and Jaak Billiet. 2006. Data Quality Assessment in ESS Round 2. Between
Wishes and Reality: Closing the Gap?. European Social Survey Web. London: Centre
for Comparative Social Surveys. Retrieved 2 August 2006
(http://www.europeansocialsurvey.org).
Särndal, Carl-Erik and Sixten Lundström. 2005. Estimation in Surveys with Nonresponse.
Chichester: John Wiley & Sons.
Sheatsley, Paul B. 1987. `Comments.' Pp. S186­S187 in `The Future Study of Public
Opinion: A Symposium', edited by Leo Bogart, Public Opinion Quarterly 51
(2: Supplement): S173­S191.
Smith, Tom W. 1995. `Trends in Non-Response Rates.' International Journal of Public
Opinion Research 7: 157­171.
Smith, Tom W. 2002. `Developing Nonresponse Standards.' Pp. 27­40 in Survey
Nonresponse, edited by Robert M. Groves, Don A. Dillman, John L. Eltinge and
Roderick J.A. Little. New York: John Wiley & Sons.
Steeh, Charlotte G. 1981. `Trends in Nonresponse Rates, 1952­1979.' Public Opinion
Quarterly 45: 40­57.
Steeh, Charlotte G., Nicole Kirgis, Brian Cannon and Jeff DeWitt. 2001. `Are They Really
as Bad as They Seem? Nonresponse Rates at the End of the Twentieth Century.'
Journal of Official Statistics 17 (2): 227­247.
STEM. 1996. Technická zpráva ­ Role vlády (ISSP 1996). (Technical Report ­ The Role
of the Government (ISSP 1996)) Prague: Středisko empirických výzkumů (STEM).
STEM. 1999. Technická zpráva ­ Sociální nerovnosti a spravedlnost (ISSP 1999).
(Technical Report ­ The Social Inequalities and Justice (ISSP 1999)) Prague: Středisko
empirických výzkumů (STEM).
Stoop, Ineke. 2005. The Hunt for the Last Respondent. Nonresponse in Sample Surveys. Hague:
Social and Cultural Planning Office of the Netherlands.
Stoop, Ineke, Jurjen Iedema, and Frans Louwen. 2000. `How Different Are
Nonrespondents?' Paper presented at the Fifth International Conference on Logic and
Methodology: Social Science Methodology in the New Millenium, Cologne, Germany,
3­6 October 2000, papers published on CD ROM.
Tuček, Milan, Lumír Gatnar and Jindřich Krejčí. 2000. Posouzení výzkumu čtenosti v rámci
multimediálního výzkumu sledovanosti médií Media projekt 2000. (Examining Research
socrev2007-1.indb 585socrev2007-1.indb 585 3.9.2007 18:41:333.9.2007 18:41:33
Sociologický časopis/Czech Sociological Review, 2007, Vol. 43, No. 3
586
on Readership in a Multimedia Survey on Media Ratings of the Media Project 2000)
Report for the Czech Publishers Association. Prague: Sociologický ústav AV ČR.
Zaletel, Metka and Vasja Vehovar. 1998. `The Stability of Nonresponse Rates According
to Socio-Demographic Categories.' Pp. 75­84 in Nonresponse in Survey Research,
Nachrichten Spezial 4, edited by Achim Koch and Rolf Porst. Mannheim: ZUMA.
socrev2007-1.indb 586socrev2007-1.indb 586 3.9.2007 18:41:333.9.2007 18:41:33
Jindřich Krejčí: Non-Response in Probability Sample Surveys in the Czech Republic
587
Appendix
Basic methodological information on selected probability sample surveys undertaken
in the Czech Republic, 1991­2005
Survey
Fieldwork
agency
Sampling method Fieldwork dates N
ISJP 1991: Social Justice STEM 2-stage stratified 1991, 26/06­14/07 810
ISSP 1992: Social
Inequalities
STEM 2-stage stratified 1992, 16/10­06/11 1101
SSVE 1993: Social
Stratification in Eastern
Europe After 1989
ČSÚ, STEM 2-stage stratified 1993, 03­04 5597
ISSP 1993: Environment STEM 3-stage stratified 1993, 20­30/11 1005
ISSP 1994: Family and
Gender Roles
Universitas 3-stage stratified 1994, 09 1024
ISSP 1995: National
Identity
Amasia 3-stage stratified 1995, 28/10­27/11 1111
ISJP 1995: Social Justice STEM Random walk 1995, 15/09­29/10 1246
ISSP 1996: Role of the
Government
STEM 3-stage stratified
1996, 20­31/10,
21/11­31/12
1100
ISSP 1997: Work
Orientations
STEM 3-stage stratified
1997, 15­30/09,
21/10­7/11,
21/11­ 31/12
1080
ISSP 1999: Social
Inequalities and Justice
STEM 2-stage stratified
1999, 18/01­01/02,
06/02­24/02
1834
ISSP 1998: Religion SC&C 3-stage stratified 1999, 18/06­27/07 1223
ISSP 2000: Environment SC&C 3-stage stratified
2000, 01/10­13/11,
21/11­20/12
1244
ISSP 2001: Social
Networks
SC&C 3-stage stratified 2001, 06/10­05/11 1200
ISSP 2002: Family and
Gender Roles
SC&C 3-stage stratified 2002, 16/09­14/10 1289
ISSP 2003: National
Identity
SC&C 3-stage stratified 2003, 26/09­19/10 1276
ISSP 2004: Citizenship SC&C 3-stage stratified 2004, 27/09­29/10 1322
ISSP 2005: Work
Orientations
SC&C 3-stage stratified 2005, 16/05­06/06 1226
Note: This information refers to the surveys represented in Figure 1. All surveys used
face-to-face interviewing. ČSÚ = Czech Statistical Office.
Source: Sociological Data Archive, Institute of Sociology AS CR.
socrev2007-1.indb 587socrev2007-1.indb 587 3.9.2007 18:41:333.9.2007 18:41:33