The Iowa Gender Wage Equity survey was designed
by Iowa Workforce Development (IWD) to assess the relationship
between wages for women and men. Based on advice from the Statistical
Laboratory at Iowa State University (Nusser and Bachan, 1999),
IWD selected a sample of 3,000 individuals using systematic sampling
on a frame developed by matching the Iowa Drivers License file
with the Wage Records file maintained by IWD, and mailed questionnaires
to the selected individuals. Because of a lower than expected
response rate and a lack of resources for a follow-up study, a
second sample of 3,000 was selected in a similar manner by IWD.
Overall, 537 individuals from the first mailing
and 523 individuals from the second mailing responded, for a total
of 1,060 respondents. There were 45 cases that were determined
to be ineligible: 6 respondents out of state, 1 deceased, 2 disabled,
10 unemployed, and 26 retired. Of the 1,015 eligible respondents,
seven returned no information, and four others did not specify
a gender and were not included in the data file for weighting,
resulting in a final sample size of 1,004. The overall response
rate for the study was 16.9 percent.
The Statistical Laboratory coded and key-entered
the data. No survey weights were calculated, because all analysis
is performed on the male and female respondents separately. The
data set was analyzed and summary tables were produced on job
characteristic differences between genders. Because the survey
asked about two jobs, it was of interest to look not only at each
job separately, but also at both of them combined. Hence, a method
for combining wages and hours worked for each respondent had to
be developed. Total hours worked was assumed to be the sum of
the hours reported for each job. Since only wage ranges were available,
the midpoints for the wage ranges (e.g. $7,500 for the range $5,000-$10,000)
for each job were used, and these were treated as actual wages.
Total wage is then the sum of these midpoints. For the highest
range ($100,000+), the value $100,000 was used since no midpoint
was available here. Only three respondents reported an income
in that highest category.
The average wage by gender was calculated for
both jobs separately and for both combined. In all cases, this
was done by dividing wages by weekly hours for each respondent,
averaging these hours-adjusted wages across respondents for both
genders, and calculating the ratio, R. Let Wgi
represent the wages and Hgi the hours worked
for individual i of gender g, where g=1 denotes
female and g=2 male. The hours-adjusted wages for respondent i of gender g are equal to ygi =
Wgi
/ Hgi. The wage ratio was calculated as
.
Some respondents reported wage information,
but no corresponding hours: 47 individuals had missing hours for
job 1 and 16 had missing hours for job 2. Whenever hours were
missing, a value of 40 hours/week was imputed for job 1 and 16
hours/week for job 2. These numbers represent the average number
of hours reported in the remaining sample. Confidence intervals
at the 95% confidence level can be calculated using the provided
standard deviations, by taking the ratio estimate and adding and
subtracting 1.96 times the standard estimated deviation 
, or
.
In order to make tables for occupations and
company type, it was necessary to compute "fractional"
respondents whenever someone reported working in more than one
occupation and/or for more than one company type. The fractions
for respondents with two jobs were calculated based on the numbers
of hours worked in each job. For instance, if someone worked 40
hours in job 1 and 20 hours in job 2, then the occupation for
this respondent becomes 67% of his job 1 occupation and 33% of
his job 2 occupation. In addition, six respondents reported more
than one occupation and/or more than one type of company within
either job 1 or job 2. For example, one person reported Sales/Service
as current occupation in job 1. In all these cases, we assumed
that this person works 50% in one occupation and 50% in the other.
The same reasoning was used for type of company. These "fractional"
occupations/company type respondents were then combined with the
remaining respondents in the calculation of the summary tables.
When hours were missing but the other information was provided
for occupation and company type, the same imputation procedure
as for the wage ratio was used.
For each question, the tabulated responses
provide estimates for the percentage of employees who selected
each of the possible answer categories. These estimates are valid
at the 95% confidence level within approximately ±4% for
females and ±5% for males. For instance, suppose that for
a certain question, 48% of the female respondents report working
in a certain occupation. One is then 95% certain that the true
percentage working in that occupation for the whole population
falls between 44 and 52%. When the results for job 2 are considered
separately, the sample size is only 137, so that the estimates
are less precise. The confidence intervals for both females and
males for job 2 are approximately ±12%.
Reference:
Nusser, S.M. and Bachan, M. Sample Design
Recommendations for the Iowa Gender Wage Equity Survey. February
17, 1999
