Introduction To Sample Surveys, Lab 6
Uploaded by
danIntroduction To Sample Surveys, Lab 6
Uploaded by
dan35100 Introduction to Sample Surveys – Lab 6
Weighting
The Parents and Teens 2004 Survey was sponsored by the Pew Internet and American
Life Project. Telephone interviews were conducted with a nationally representative
sample of 1100 teens 12-17 years old and their parents living in continental United States
telephone households. The interviewed sample was weighted to match national
parameters for both parent and child demographics. The parent demographics used for
weighting were: sex; age; education; race; Hispanic origin; marital status and region
(U.S. Census definitions). The child demographics used for weighting were gender and
age.
Weighting was accomplished using Sample Balancing, a special iterative sample
weighting that simultaneously balances the distributions of all variables using a statistical
technique called the Deming Algorithm. Weights were trimmed to prevent individual
interviews from having too much influence on the final results. The use of these weights
ensures that the demographic characteristics of the sample closely approximate those of
the national population.
Open the Parents and Teens data set in the UTS Online folder.
You can see that there are 185 variables, one of which is a weighting variable.
6.1. Find frequency tables for the census region (cregion: variable 11), the parent’s
education (EDUC: variable 45) and the parent’s race/ethnicity (RACE: variable 47).
When you find these tables use the method of clicking Paste rather than OK in the
dialogue box. This will open the Syntax window and paste the commands in the Syntax
window. Select Run > All to get your output.
Make a note of the percentages in each category in your tables.
How many total respondents do the tables say are in the sample?
cregion EDUC RACE
Northeast 18.8% None/1-8 1.2% White 78.5%
Midwest 23.7% HS Incomplete 8.2% Black/AA 12.6%
South 36.7% HS Graduate 32.5% Asian/PI 3.1%
West 20.8% Technical Sch 3.6% Mixed 1.1%
Some College 24.6% Native Am 2.0%
College Grad 18.0% Other 1.7%
Postgrad 11.6% Don’t Know/Refused 1.0%
Don’t Know/Refused 0.2%
There are 2717 respondents recorded in this table.
35100 Introduction to Sample Surveys, lab 6 1
6.2. These tables have actually been produced using weighting. To find out what the
unweighted figures would have been, type WEIGHT OFF. at the start of the Syntax
window with a blank line after it. Then run all the commands again.
Compare the percentages with those you found before. What are the main differences
between the weighted and unweighted percentages you have found?
What is the total number of people who were interviewed?
There are now a smaller proportion of Northwest respondents and western respondents
and a larger proportion of Midwest respondents. The proportion of people with lower
levels of education has decreased and the proportions of those with higher levels of
education have increased. The proportion of “White” respondents has increased, while
the proportions of other respondents have gone down.
There are now 1100 respondents.
6.3. If you wanted to reproduce the weighted frequency table for census region, you
would need to type in the Syntax window the following command.
WEIGHT by weight .
FREQUENCIES VARIABLE=cregion .
Try to find the unweighted and weighted frequency tables for questions k5a to k5d and
see if there were any marked differences between the weighted and unweighted
percentages. Summarise these differences below.
The response proportions are slightly lower in the weighted sample (86.7%) than in the
unweighted sample (88.3%). For QK5a – QK5c, the proportion of people who need help
is slightly lower in the weighted sample than in the unweighted sample. In QK5d, the
proportion of people who need help is lower in the weighted sample than in the
unweighted sample.
35100 Introduction to Sample Surveys, lab 6 2
6.4. Suppose that we would like to recalculate the weights based on only the variable
cregion. Suppose that the population estimates for each region are:
Northwest: 678
Midwest: 759
South: 1077
West: 486
Using the procedure discussed in the lecture, recalculate the weights. (NOTE: In the
lectures, we created weights to distribute the interviews in each slot to match the
distribution of visitors in each slot. Here we want to distribute the responses from each
region to match the above population estimates for each region.) What weights are given
to each region? (Evaluate your answer to 3 decimal places)
Northwest weight = 678/186 = 3.645
Midwest weight = 759/297 = 2.556
South weight = 1077/401 = 2.686
West weight = 486/216 = 2.25
6.5. Create a new column using Transform > Recode into Different Variables to
recode the cregion variable into a new weight variable called newweight. Repeat Q6.1
using the new weights (so use the command WEIGHT by newweight).
cregion EDUC RACE
Northeast 22.6% None/1-8 0.6% White 85.4%
Midwest 25.3% HS Incomplete 3.7% Black/AA 9.2%
South 35.9% HS Graduate 27.0% Asian/PI 2.3%
West 16.2% Technical Sch 2.9% Mixed 0.7%
Some College 26.9% Native Am 1.1%
College Grad 23.2% Other 0.7%
Postgrad 15.5% Don’t Know/Refused 0.7%
Don’t Know/Refused 0.2%
There are 3000 respondents recorded in this table.
Note that weights are usually calculated using more than one demographic variable.
