- Teaching Resources
- The Council
GUN_CONTROL6G: Exercise Using SDA and the Field Polls for Trend Analysis by Breaking the Sample into Subgroups
Author: Ed Nelson
I’m attaching the following files.
Note to the Instructor: This exercise is a continuation of the previous exercise (GUN_CONTROL5G). In that exercise we used three Field Polls to look at trends in opinion on gun control. In this exercise we’re going to break the samples down by sex and look at the trend for men and women separately. You have permission to use this exercise and to revise it to fit your needs. Please send a copy of any revision to the author. Please contact the author for additional information.
Goals of Exercise
The goal of this exercise is to provide an example of trend analysis. Using the same three Field Polls from the previous exercise, we will break the samples down by sex and look at the trend for men and women separately. The exercise also provides an introduction to SDA (Survey Documentation and Analysis), an online statistical package.
Part I—Using the Field Polls
The Field Polls are archived at UCDATA, an archive of social science data at UC Berkeley. To access the Field Polls click on this link. Notice that how you access the Field Poll data depends on where you are accessing the data from. If you are on a campus of the University of California or the California State University, you would use the links for UC and CSU affiliates. Otherwise you would use the links for “other researchers, non-UC/CSU academic institutions, or general public.”
UC and CSU affiliates have access to all Field Polls. A Field Poll is typically available within approximately 90 days of its completion. However, polls are embargoed for two years for other researchers and the general public. In other words, polls are not available for two years after the data are collected for other researchers and the general public. This exercise is going to access the polls as the general public so we will only be able to access Field Polls conducted no less than two years ago.
Under “Other researchers, non-UC/CSU academic institutions, or general public” you should see three links.
Click on the link to search the Field Polls. You should see two tabs – “Choose Collections/Datasets to Search” and “Search Options and Results.” The first of these tabs is the default and should already be selected. Check the boxes representing the time periods you want to search. For this exercise, click the boxes for the 1990’s, 2000’s, and the 2010’s.
Now click on the other tab for “Search Options and Results.” Notice that there is an option for “Search Techniques Help.” Click on this link and read the help page. Notice particularly that the wildcard is the asterisk (*). If you were to enter gun* you would be searching for all words starting with gun which would include gun and guns. Notice also that below the keyword box you can choose whether you want to search for variables that include “all” search words or variables that include “any” search word. The default is to search for variables that include “all” search words. We’re going to use the default in this exercise. Now enter the keyword gun* in the search box and click on the search button.
One of the questions that the Field Poll has asked in a number of polls is the following: “What do you think is more important – to protect the right of Americans to own guns, or to impose greater controls on gun ownership?” The search results should show you these polls plus all the other polls that have the words “gun” or “guns” in one of their questions. You’ll be able to identify them by the question that is listed on the search results. You should find five polls that included this question. Click on the “View” button for each of these polls to see how respondents answered this question. However, a word of warning. Don’t pay attention to the percents that you see when you click on “View. Sometimes these percents include those who weren’t asked the question. That’s what “Not Applicable” means. It’s common in polls to ask some questions of a random half or random third of the sample. It’s one way of including more questions in the poll without adding to its length. Additionally, these are the unweighted percents. Poll data typically use a weight variable so that the results will better represent the population from which the sample is selected.
We’re going to use the same three polls that we used in the previous exercise.
Part II – Analyzing the Data
Now that we have identified the three polls that we’re going to use, we need to analyze the data. Look for “Cal Poll 1301: February 5-17,2013.” Clicking on the study name will open SDA. On the left you will see a list of the question categories for this survey. Look for “Guns” and click on the arrow to the left of the category title. The variable name for the question we will be using is q13. (Note that, for this Field Poll only, the variable names must be in lower case.) Then look for the category called “Demographics and Party Identification.” Click on the arrow to the left and scroll down to the bottom. You should see a variable called weight (must be lower case). This is the variable that will weight the cases so they better represent the population from which the sample was selected.
Now we’re ready to analyze the data. We’re going to look at trends for males and females separately. Enter q13 (opinion on gun control) in the Row box, q110 (sex) in the column box, and weight in the Weight box. Click on “Run the Table” and you will see the weighted percentages broken down by sex. There are three response categories. Some men (42.9%) said that the “right to own guns” was most important. Other men (51.8%) said that it was more important to “control gun ownership.” And still other men (5.3%) had no opinion. Women responded quite differently. Some women (26.4%) said that the “right to own guns" was most important, while other women (68.5%) said it was more important to “control gun ownership” and still other women (5.1%) had no opinion.
The next poll we’re going to use is “Cal Poll 0203: April 19-25, 2002.” Follow the same procedures that you used in the 2013 survey. Find the variable name for opinion on gun control and enter it in the Row box. Enter the variable name for sex in the column box. This time the weight variable has already been entered for you.
There’s a complication with this survey that we have to deal with. Prior to 2006 the Field Poll was a random sample of all adults in California. From 2006 on it was a random sample of registered voters in California. When we used the 2013 Field Poll we knew that all the respondents were registered voters because of the way the sample was selected. But the 2002 poll and all polls prior to 2006 will include both registered and non-registered voters. To make the polls comparable we need to select out only the registered voters for analysis. Look at SDA again. You’ll see a box called “Selection Filter(s).” This is where you tell SDA to select out particular cases. In the April, 2002 poll the variable that tells us who is registered is q18 and registered voters are coded 1 through 4. (We’ll exclude the few respondents who said they didn’t know if they were registered.) To select out these cases, enter q18(1-4) in the “Selection Filter(s)” box. This should leave you with 653.1 cases.
But there’s another issue to consider. This time only a random half of the respondents were asked the question so “Not Applicable” includes about 50% of the sample. That means you will need to recompute the percents so they exclude those who weren’t asked the question. Add up the number of men who replied right to own guns, control gun ownership, and those who had no opinion. This should add to 151.0. Then do the same thing for women. Now recompute the percents using 151.0 as the base for the percents for men and 175.3 for women. This means that for the men you would divide 72.4 by 151.0 which will equal 0.479 and convert this to a percent (47.9%). For women you would divide 57.0 by 175.3 which will equal .325 and convert it to a percent (32.5%). Do this for all three responses.
The final poll we’re going to use is “Cal Poll 0003: June 8-18, 2000.” Follow the same procedures that you used previously. Find the variable name for opinion on gun control and enter it in the Row box. Now find the variable name for sex and put it in the Column box. The weight variable has already been entered for you. You will need to select out the registered voters again. In the June, 2000 poll the variable that tells us who is registered is VARI_17. To select out these cases, enter VARI_17(1-6) in the “Selection Filter(s)” box.(As before, we’ll leave out those who declined to state their registration status and those who said they didn’t know.) This should leave you with 583.9 cases.
In this poll only a random half of the respondents were asked the question so “Not Applicable” includes about 50% of the sample. That means you will need to recompute the percents for both men and women so they exclude those who weren’t asked the question. Add up the number of men who replied right to own guns, control gun ownership, and those who had no opinion. This should add to 144.1. Now recompute the percents for the men using 144.1 as the base for the percents. This means that you would divide 71.8 by 144.1 which will equal 0.49.8 and convert this to a percent (49.8%). Do the same thing for women. Do this for all three responses.
Part III – Looking at Trends over Time
We now have three polls that asked the same identical question. The 2013 poll was a sample of registered voters while the 2002 and 2000 polls were samples of all adults in California. We made them comparable by selecting out the registered voters in the 2002 and 2000 polls so the data were comparable. In the 2013 poll the gun control question was asked of all respondents while in the 2002 and 2000 polls it was only asked of a random half of the respondents. We adjusted for this difference by recomputing the percents for the 2002 and 2000 polls so the base included only those who were actually asked the question. These were important steps in our analysis of the data. We wanted to make sure that our data were comparable for the three polls.
Now let’s create a table that makes it easy to visually compare our adjusted percents for men and women. Any word processor will construct tables for you. In Word, click on insert and then the drop-down arrow below “Table” and then “Insert Table.” Tell Word how many rows and columns you want in your table and Word will insert the table into your document. We want to have seven rows and seven columns. To merge cells select the cells you want to merge and then click on “Merge Cells.” Now label the rows and columns. Your table should look like this.
We can edit our table to make it look nicer but that’s not necessary. Fill in the recomputed percents from your analysis in Part II.
Write a paragraph or two describing the trend over time in opinion on gun control for men and women separately. Be sure to include in your essay a description of the steps you took to make sure than the information is comparable from poll to poll. Consider whether there has been more change for men or for women. Are men or women more likely to favor controlling gun ownership? Does this vary by time period?
 Even if you are a student or faculty member at one of the California State University campuses you must be on campus to use the links for “UC and CSU affiliates.” That’s because access is IP authenticated. Additionally, only those CSU campuses that subscribe to the social science data bases (e.g., Inter-University Consortium for Political and Social Research, Roper Center for Public Opinion Research, The Field Poll) may access recent (less than two years old) Field Polls.
 Metadata are data about data that describe the variables in the various polls. Since 2008 the metadata include the exact question wording for each variable. This makes it easier to find what you are looking for. Before 2008 the metadata were short descriptions of the questions. That makes it more difficult to search before 2008.
 This is the Field Poll that we used in the first four exercises in this series.
 You are probably wondering why the number of cases is not a whole number. The reason is that weighting creates partial cases. Don’t worry about this.