Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 60
60 Standardized Procedures for Personal Travel Surveys
4.2.14 S-5: Specification of Sampling Error Requirements
Frequently, RFPs specify that the required sample must provide no more than, say, ±10% error
with 95% confidence in something such as a trip rate. Generally, this appears to be specified with
little understanding of what it means. It would be reasonable to question whether 10% error is
acceptable compared with 5% error and whether the significance level should be set to 90% or
95%. Also, the error is almost always specified for trip rates, while the data will be used for much
more than trip-rate estimation. The implications of a particular error level for trip rates on esti-
mation of such elements as mode choice or network volumes are largely unknown.
The first issue that needs to be addressed is to determine an appropriate specification of the level
of error and confidence level to be used in designing samples. This issue could be researched by
using existing data sets. In this case, variations in data caused by differences in survey protocols,
firms, survey instruments, etc., would be irrelevant. It may be most useful to present graphs show-
ing the effects of changing each of the error levels and the confidence level so that the implications
of each can be seen. This should, ideally, be done using actual computations of sampling error from
recent surveys. The implications of the level of error can be investigated by examining simple trip-
production models and showing the implications in terms of ability to distinguish statistically
between the trip rates for different population sub-groups.
The second issue is to determine the implications for other variables that may be estimated from
the data of setting an error level on trip rates. To do this, one would need to select certain other
variables of interest--the proportions of trips by mode and purpose, the average trip length by pur-
pose, trip rates by purpose, average household size, average vehicle ownership, etc.--and estimate
the sampling errors on these attributes. These would need to be related to the sampling errors for
the overall trip rates to show how the sampling errors on the other attributes relate to the trip-rate
sampling error. If there is insufficient variability in the overall error of trip rates, it may be neces-
sary to sub-sample from some existing surveys since the sub-samples will have much larger sam-
pling error for all characteristics.
The third issue is to investigate the potential to use other attributes, such as mode shares, for the
design sampling error. Existing data sets could be used to determine the error properties of such
attributes as mode shares and possibly other attributes like average trip lengths by purpose.
If the attribute on which the sampling error is specified is changed, then a different type of sam-
pling will be required to achieve the desired sample. This would require investigation of what would
be required and how it could be attained. Existing data sets could be used for this--for example, a
secondary data source like Public Use Microdata Sample (PUMS) could be sampled to replicate
the procedure that would need to be used.
4.2.15 S-6: Development of Default Variances
Estimation of error requires an estimate of the variance of crucial variables. One of the issues
that has made sampling strategies relatively simplistic in household travel surveys is the lack of
information on variances for those variables that are normally considered crucial in transporta-
tion planning analysis. This has implications on all aspects of sampling because the error levels
are determined by the variance; hence, sample size and stratification procedures are also deter-
mined by the variance. In the absence of information about the variance, survey designers either
assume constant variances across all strata in a sampling scheme or make some other working
assumption that will allow sample size calculations to be made.
Default variances could be used to determine appropriate sample sizes and other issues in the
absence of actual local data on these values. They could also be used subsequently to assist in
