Vinnumarkaður - 01.09.1996, Blaðsíða 27

Synopsis of methods and concepts
25
This means that in 19 surveys out of 20 the mean unemploy-
ment in 1995 would have been estimated between 6,200 and
8,200.
Estimates with relative standard error exceeding 20% are
marked with an asterisk (*), cf., e.g., Table 2.4.
Non-sampling errors. There are three categories of non-
sampling errors: Coverage errors, non-response errors and
other errors.
Coverage errors. The sampling frame consists of persons
with domicile in Iceland. A certain portion of these reside
abroad for more than 6 months, however, because of work or
study. Only a small number of these people actually register
their foreign residence at the National Register. This results
in an over-coverage of the survey population and a corre-
sponding bias in estimates. All estimates have been corrected
on basis of the sample to remove this bias (cf. Table 2.4).
Under-coverage errors havenotbeen detected to any degree.
Non-response errors. Males are more difficult to reach than
females, both because of absence from home or non-contact.
Y oung people have ahigher non-contact rate than older people
and inhabitants of the capital region are more difficult to
contact than others. Refusals are more prevalent among
women, inhabitants of the capital region and older persons.
Response rates are, however, higher for women, people
outside the capital region and middle-aged persons as these are
less likely to be away fromhome or not to be found (Table 2.5).
To counter possible non-response bias the results have
been weighted by sex and age group information from the
National Register. No use was made of residence data, as these
are less reliable than other data.
It is more difficult to adjust for non-response bias directly
related to the subject matter of a labour force survey. Certain
occupational groups are more difficult to contact than others,
e.g., fishermen. From November 1993 this has been dealt with
by asking close family members of the sampled individuals to
answer the questionnaire. The number of proxy answers is,
however, less than one percent of all responses.
In most cases the item non-response has been corrected for
in order to preserve consistency in totals as well as to counter
bias resulting from non-response. Two main methods have
been used. Firstly, a predicted answer has been deduced frorn
the answers of similar respondents. Missing data on working
hours were, e.g., replaced by regression coefficients from a
regression analysis using sex, full-time/part-time employ-
ment and occupational group as independent variables. Sec-
ondly, the most probable response category has been deduced
from other responses by the respondent either in the same
survey or other surveys in which he or she participated.
Other errors. Other errors are classified into three categories
in this context: Interviewer errors, processing errors, and
designerrors.
Interviewererrors are those resulting from wrong recording
of answers, omission of questions, wrong routes in the
questionnaires or the rephrasing of questions by an inter-
viewer. These errors were more pronounced before the com-
puterisation of the interviewing process.
These errors have mainly been dealt with by hiring only
experienced interviewers, by interviewer training and special
meetings with interviewers before each survey.
Processing errors. Before computerisation of the survey
in November 1992, errors sometimes happened when data
from paper questionnaires were entered into computer files.
Coding of certain open questions, especially regarding occu-
pation, economic activity and educational level is also prone
to errors.
These errors have mainly been dealt with by using experi-
enced coders, by careful editing of these variables and system-
atic searches for inconsistencies.
Design errors. During the first three years, the question-
naire has undergone changes, although the main body of
questions has stayed the same. This is especially true with
regard to the phrasing of certain questions and their sequence.
During the period many questions have also been added in
order to conform more fully to the requirements of the EEA
agreement. Bias related to these sources is not known. The
following are three examples of known design errors relating
to wording of questions, the frequency of the survey and
annual estimates and the rotation scheme.
Wording ofquestions. Until April 1994 the sub-category
“unpaid family worker’’ was distinguished in the question-
naire by the wording “works at a family business/with spouse,
parents or children”. This led to confusion as to who in the
family business are self-employed and who are unpaid family
workers, notwithstanding the validity of this distinction in a
modern society. Since November 1994 the sub-category was
rephrased to “works at a family business without pay” and
the interviewers were given special instructions as how to
make the distinction. This resulted in a “drop” in the number
of unpaid family workers from approximately 2.6% to 1.1%.
Frequency ofthe survey andannual estimates. As the Iabour
force survey is only conducted twice each year, estimates of
totals which are subject to seasonal variations can differ from
the true totals. Using registered unemployment in April and
November 1975-1995 as compared with the annual averages
in this period as a benchmark, the estimates of the annual total
using only these two months seem to be somewhat conserva-
tive on the average and approximately non-biased during the
period from 1988-1995 (cf. Figure 2.1).
Renewal ofthe sample. As each rotation group is selected
by a simple random sampling with equal sampling fraction
across age cohorts, the two youngest age groups are
underrepresented in the sample. This results in biased esti-
mates of the total labour force participation rate as these age
cohorts have a lower labour force participation than the
average. This has been dealt with by using a weighting scheme
with age groups 16, 17, 18-19 and thereafter at five-year
intervals. As mentioned above, from November 1994 the
sampling scheme was altered in order to ensure proportionate
representation of these two age cohorts.
3.1.4. Concepts
Main activity. In four tables (Tables 6.58-6.61) the data are
classified according to main activity status, which differs from
the ILO recommendations. Employed persons in these tables
are those whose usual hours of work are 12 hours or more per
week, whereas unemployed people are those who are actively
seeking employment, with the exception of students who are
only looking for a part-time job. Economically inactive per-
sons are classified according to their reported status.