Constructing/transforming Variables
• Still preliminary to data analysis (statistics)
Would fit comfortably under
Measurement
A bit more advanced is all
• All the earlier material about operationalization (importance of it, difficulty of doing well, need to think about and assess reliability and validity) apply
How (and why) change variables?
• Collapse codes
• Reverse coding
• Combine items
• Standardize items
Collapse codes
• Example: Party identification
1 = SD, 2 = WD, 3 = ID, 4 = Ind,
5 = IR, 6 = WR, 7 = SR
Collapse to 1 = Dem (1-2 above),
2 = Ind (3-5), 3 = Rep (6-7)
Collapse codes (cont.)
• Why collapse codes?
For theoretical reasons
(In a given instance) we think only
direction of partisanship matters
To combine small categories
Another option is to delete those cases
To get a reasonable # of categories
Age 18, 19, 20…97 = 80 categories
Reverse coding
• Examples: 1 = Disagree, 2 = Neutral, 3 = Agree Reverse so 1 = Agree, 2 = Neutral, 3 = Disagree
1 = Low…10 = High Reverse so 1 = High…10 = Low
Reverse coding (cont.)
• Why reverse coding?• Questions are reversed in surveys Example: from homework, next slide Indicators have “opposite” meanings Example: unemployment (up is “bad”); increase in income (up is “good”)• With reversal: It’s often easier to interpret items Combining items make sense
Tolerance questions
• Members of the [least-liked group] should be banned from being president of the U.S.
Disagree is the “tolerant” response.
• Members of the [least-liked group] should be allowed to teach in public schools.
Agree is the “tolerant” response.
Combine items
• Examples can be complex (we’ll see later)
• Simple example:
Count the number of correct, or
tolerant, or liberal, or … responses
As is Knowledge and Tolerance scales
(in homework)
Combine items (cont.)
• Why combine items?• Multi-variable items are usually more valid
Many concepts—e.g., type of election
system, tolerance, knowledge—are hard
to measure with a single question or indi-
cator (they contain multiple components).
In a combined measure, we can include items
that measure all of these components
Combine items (cont.)
• Combined (multi-variable) items typically increase reliability as well
Random error that affects individual
items is averaged out
• Combining items also yield more refined measurement
Simply put, we get more categories
Ex: Pro-business support
• Conceptual definition is simple: how favorable toward business are members of Congress?
• More specifically, rank U.S. House members by their favorability toward business.
• How might we do this?
Standardize items• Why standardize items?
• To account for different numbers of base items.
• To compare or combine items measured on different scales altogether.
Save this for later.
Standardize items
• Examples
Percentaging (e.g., 90% is “equivalent”
despite different length tests).
Making measures comparable by
deflating for changing bases (e.g.,
increased media coverage over time).
Media coverage over time: More pages & more periodicals
Deflating for total volume matters
• Downward trend in coverage of auto safety masked by increasing media volume.
A reminder
• Composite (multi-variable) measures are not always better (more valid/reliable).
• Results can be artifact of how you construct your variables.
Real world example of a composite (better?) measure
• Likelihood of voting (in election polls)
Need to estimate because many who are
interviewed will not vote
People won’t/can’t estimate own behavior• Pollsters use information about past voting,
whether registered, interest in the race, etc.
But: polls vary (in part) because different
pollsters use different sets of questions
Why change variables?Additional reasons
• To change the nature of the variable
Example: log transformation (age often
done this way)
• Create new variables
“Just” combining items again, but it can
be very complex
Example: next slide
Example: Clarity of responsibility for government decisions (Powell)
• Idea is that in some instances it is easy to assign credit or blame for what a government does; other times not
• Powell builds an index using measures of:
Presence of minority government
Whether there is bicameral opposition
Number of political parties in the legislature
Degree of cohesion among the parties
Strength of committee chairs in the legislature
The text on combining/altering variables in SPSS
• Text talks about Recode and Compute operations for simple purposes
Collapsing Additive index (e.g., # of yes answers) Useful but far from all that you can do• Mentions transformations Refers to “a dizzying variety of complex transformations” possible” (text)
• Comment/advice:
If you want to do it, you almost certainly
can do it in SPSS—and you can
probably do it quite easily.
Ex: Make individuals (in a survey) a 1 if
var 5 + abs. value of var 6 + 2(var17) +
log(var 19) is greater than 24.5 OR if var 3
equals 4; otherwise make them 0
• Final note on SPSS
When you recode:
Save in a new variable almost always
Checking after each operation
• More on changing variables in the lab
sessions