Ferðavenjur Íslendinga - 01.02.1998, Blaðsíða 29

Ferðavenjur íslendinga 1996
27
tourist industry, on models from neighbouring countries
where similar surveys had already been conducted, on the SI
statistics on tourist accommodation, and on other sources.
Data collection
The survey was conducted by telephone. Every sampled
participant received a letter in advance explaining the back-
ground and aim of the survey with a request for his or her
cooperation. These people were subsequently contacted by
telephone, chiefly in the evening and during weekends.
Telephone numbers were obtained from the Post and Tel-
ecommunication Administration. The survey was conducted
with the help of the interviewing program BLAISE, which
greatly facilitates the whole process of making a survey, both
in terms of execution and processing of the data.
Survey period
The survey covered the year 1996 and was conducted in three
stages. The first stage extended to trips taken in the period
January-May and took place at the end June. The second
stage covered trips taken in the months April-August and this
part was performed in September. Finally, the third stage
took place in January 1997, covering trips taken in July-
December. It was deemed advisable not to cover longer
periods in each stage so that the respondents would easily be
able to recall their trips. The periods were deliberately made
to overlap in order to make it possible to produce figures for
both three-month and four-month periods. Eurostat requests
quarterly statistics whereas this publication shows figures for
four-month periods. For Iceland it is more practical to em-
ploy four-month periods so as not to split up the major tourist
seasons, such as Easter (in March and April) and summer (in
June and July).
Sample and response rates
The sampling frame in each stage of the survey included all
Icelandic and foreign citizens aged 16-74 who were on the
National Register of Persons and had their domicile in
Iceland during the survey period. Each sample was made up
of 1.200 persons picked by random sampling from the
National Register. For the purpose of including children
under the age of 16, mothers were asked about the tourism
patterns of their children, or alternatively, single fathers were
asked those questions. The mothers were preferred as
respondents to these questions since they were considered
more likely than the fathers to know in some detail about their
children’s travels. In the first stage of the survey the final
sample, including grown-ups and children, tumed out to be
1,600 individuals, in the second stage this number was 1,681
and 1,643 in the third one. The combined sample size in all
three stages was 4,812 individuals and the response rate was
87%. A more detailed account of the response rates is found
in Summary 20.
Reliability
Errors in a sample survey fall roughly into two major catego-
ries, sampling errors and non-sampling errors. The follow-
ing section deals mainly with those types of errors that are of
significance to the processing of data in this particular
survey.
Sampling errors
Surveys based on samples have an inherent degree of uncer-
tainty since the sampled individuals have been drawn at
random from a particular sampling frame such as the Na-
tional Register of Persons. The uncertainty stems from the
fact that the sample may not be an exact reflection of the
population in question. The random nature of this uncer-
tainty makes it necessary to calculate confidence limits for
the estimates. Summary 21 shows 95% confidence limits for
the number of individuals in the Statistics Iceland tourism
survey. If, for instance, the number of tourists in the period
May-August aged 45-64 years is estimated at 42,500, the
confidence limit for the number which comes closest to this
estimation is +/- 4,600. This means that there is a 95%
probability that between 32.900 and 42.100 Icelanders aged
45-64 took some trip in the period May-August. If the size
of the estimate as less tlian 4,000, the relative standard error
will exceed 20%. Estimates, percentages and averages for
groups smaller than 4,000 are marked with an asterisk (*).
Summary 22 shows 95% confidence limits for the number of
trips and confidence limits for each category are calculated in
the same manner as described above.
Non-sampling errors. There are three different categories
of non-sampling errors: Coverage errors, non-response er-
rors and other errors.
Coverage errors. Coverage errors are caused by two things.
On one hand, the register used for the sampling, i.e. the
sampling frame, may not be exhaustive and, on the other
hand, it may include individuals or units that do not belong
to the population being investigated. These types of errors
are referred to as undercoverage and overcoverage errors
respectively.
As described above the sample for the tourism survey was
drawn from individuals aged 16-74, domiciled in Iceland
according to the national register. The national register,
however, includes a considerable number of people studying
or working abroad for longer than six months at a time. Only
a portion of this group is registered as residing abroad in the
national register. This will result in a considerable bias so
that totals are overestimated unless this group is subtracted
from the total population.
For this reason, all totals for number of trips and number
of tourists are based on the mean population in 1996 as
calculated by Statistics Iceland and corrected by subtracting
the number of persons presumed to be residing abroad and
domiciled in Iceland according to indications supplied by the
tourism survey. Summary 23 shows a summary of the mean
population in 1996 by age and residence as estimated by
using the above methods. Undercoverage in the national
register, however, has not been detected to any degree.
Non-response errors. In all surveys there is a risk that
results become biased because non-response is not evenly
distributed among the groups examined. The most common
reasons for non-response are refusals, difficulties due to
illness or disability, absence from home during the time of
survey or non-contact because addresses or telephone num-
bers of sampled individuals could not be found. In general,