Survey Methodology 3. Response Rates
This section
describes the computation of the response rates for the 2002 National
Transportation Availability and Use Survey. The overall response rate is the
product of the screener interview response rate and the extended interview
response rates. The screener and extended interviews were computed in
accordance with the standards defined by the Council of American Survey
Research Organizations (CASRO, 1982). The response rate was also computed using
the survival method (Brick, Montaquila, and Scheuren, 2002) for reference. The
weights for the survey data files were adjusted using the CASRO definition as
described in section 4.
To compute the
screener response rate, the study classified each screener disposition code as residential, nonresidential or undetermined,
as Table 3.1 shows. To classify these cases, a variable, RESDFLG (residential
flag), was created with values of YES (residential), NO (nonresidential) and
UNK (undetermined or unknown). Table 3.1 shows the classification of the
screener disposition codes and the number of cases for the survey. Notice that
the only screener disposition codes that are considered undetermined are NA and NM (ring no answer, and answering machine).
This three-part breakdown of cases for the RESDFLG variable is presented in
Table 3.2.
The number of completed screener
interviews (cases coded as CA, CB, CD, and CS) is 10,327, as defined in the
first four rows in Table 3.1.
The screener response rate (unweighted) is computed as
![If you are a user with a disability and cannot view this image, please call 800-853-1351 or email answers@bts.gov for further assistance.](https://webarchive.library.unt.edu/eot2008/20090513061247im_/http://www.bts.gov/programs/omnibus_surveys/targeted_survey/2002_national_transportation_availability_and_use_survey/public_use_data_files/images/image004.gif)
Different definitions exist for the
screener response rate. These definitions depend on the way the number of
residential cases among the undetermined is estimated.
3.1 Computing a CASRO Response Rate
In CASRO, the proportion of undetermined numbers
that are residential numbers is estimated using the observed residential and
nonresidential cases in the sample. This proportion is computed as
![If you are a user with a disability and cannot view this image, please call 800-853-1351 or email answers@bts.gov for further assistance.](https://webarchive.library.unt.edu/eot2008/20090513061247im_/http://www.bts.gov/programs/omnibus_surveys/targeted_survey/2002_national_transportation_availability_and_use_survey/public_use_data_files/images/image006.gif)
Thus, CASRO assumes that 40.18
percent of the undetermined cases (NA and NM) are residential.
Using the proportion based on the CASRO formula,
the screener response rate is then
.
The extended response rate
(unweighted) is computed as
.
Table 3.3 shows the number of completed cases by
extended interview disposition code. A total of 5,019
out of 5,755 extended interviews were completed.
Therefore, the extended response
rate is
![If you are a user with a disability and cannot view this image, please call 800-853-1351 or email answers@bts.gov for further assistance.](https://webarchive.library.unt.edu/eot2008/20090513061247im_/http://www.bts.gov/programs/omnibus_surveys/targeted_survey/2002_national_transportation_availability_and_use_survey/public_use_data_files/images/image012.gif)
and the overall CASRO response rate is
OVERALLRR = SCRRR * EXTRR
= 0.642492 * 0.872111 = 0.560324 = 56.03%.
3.2 Computing a Survival Analysis Response Rate
The survival
analysis method is used to compute response rates when there is an additional
effort made to resolve the residential status of the cases coded as NA and NM
(for example, by making repeated calls to a sample of these cases to determine
their status). If there is no such effort, the survival method will produce the
same results as the CASRO formula (Brick, Montaquila, and Scheuren, 2002). If
there is an additional effort, the screener response rate is usually higher
than the CASRO rate.
This
additional effort was done for another RDD study, the very large National
Survey of American Families (NSAF), an RDD survey with approximately 46,000
completed interviews. This allowed the computation of a response rate for the
Transportation Availability and Use Study based on the NSAF survival method.
NSAF was very similar to this survey in the following ways:
- A national survey;
- Conducted during the same time; and
- The proportion of observed residential cases to total cases (cases determined
to be residential/total cases) was almost identical.
In NSAF, an
additional effort was made to resolve the residential status of the NA and NM
cases by making repeated calls to a sample of these unknowns. Therefore, in
NSAF the response rate and residential rate were computed using the survival
method. The observed residential rate is a population value; therefore it
should be the same for both this and the NSAF surveys. For this reason, one can
apply the response rate computed in NSAF to this survey.
The study team
proceeded as follows when applying the NSAF rate to this survey:
- Using the NSAF residency rate, and the formula for the residential rate,
the study computed the estimated proportion of NA and NM cases that were residential.
- The study team then used the proportion of NA and NM cases that were considered
residential from (1) in the screener response rate formula to compute the
screener response rate.
The proportion
of NA and NM cases that were residential is computed by solving for p in the formula for the residential
rate. This formula for the residential rate is
![If you are a user with a disability and cannot view this image, please call 800-853-1351 or email answers@bts.gov for further assistance.](https://webarchive.library.unt.edu/eot2008/20090513061247im_/http://www.bts.gov/programs/omnibus_surveys/targeted_survey/2002_national_transportation_availability_and_use_survey/public_use_data_files/images/image016.gif)
The NSAF residency rate was computed as 0.391, so solving for p and substituting the number of cases,
the value of p is
.
Using this p,
in the screener response rate formula, the screener response rate based on the
survival analysis is
.
The overall response rate is
then calculated as
.
That is an increase of 1.55% compared to the CASRO rate. Table 3.4
summarizes the response rates. The weights for the survey data files were
computed using the CASRO definition.
References
Council of
American Survey Research Organizations. (1982). On the definition of response rates (special report of the CASRO task
force on completion rates, Lester R. Frankel, Chair). Port Jefferson, NY: CASRO.
Brick, J.M., Montaquila, J., and Scheuren, F.J. (2002). Estimating residency
rates for undetermined telephone numbers, Public Opinion Quarterly, Vol.
66, 18-39.
|