6. +
Operationalization
Process of connecting concepts to
observations
Developing procedures to measure variation
in variables
Process of specifying the operations that will
indicate the value of cases on a variable
7. +
Concepts Variables |Dimensions Indicators
Binge Drinking Frequency of heavy
episodic drinking
“How often within the
past two weeks did you
consume 5 or more
drinks containing
alcohol in a row?”
Poverty Subjective poverty
Absolute poverty
“Would you say you
are poor?”
Family income/Poverty
threshold
Socioeconomic
Status
Income
Education
Occupational prestige
Income + education +
prestige
10. +
Nominal
Variables have two or more categories
No numeric scales but assigned a number
Groups such as
Male-female
Study in quiet room vs noisy room
12. +
Ordinal
Rank order points on a scale
Intervals between items are not known or are
not equal
Examples:
2, 3, or 4-star restaurants
Ranking TV shows by popularity
13. +
Within
the last year about how
have used…..
Did
not
use
1X
per
year
6X per
year
1X per
month
2X per
month
1x
per
week
3X
per
week
5X
per
week
Every
day
Tobacco
Alcohol
Marijuana
Cocaine
Amphetamines
Sedatives
Other illegal drugs
15. +
How do you think your close
friends feel (or would feel) about
you……
Don’t
disapprove Disapprove
Strongly
disapprove
Trying marijuana once or twice
Smoking marijuana occasionally
Smoking marijuana regularly
Trying cocaine once or twice
Taking cocaine regularly
Trying LSD once or twice
Taking LSD regularly
16. +
Ratio scale
A true zero point – absence of the variable
Zero on weight means no weight
Can form ratios: 10 pounds is twice as heavy
as 5 pounds
Use more sophisticated statistical tests for
ratio and interval scales
You can always transform ratio-level
variables into lower-level variables
17. +
Having observed two youngsters take a nasty spill while fetching water,
Research Smith wondered if a warning message to “be careful while
climbing the hill” might help avert this sort of mishap.
He also wondered if males were more likely to fall first, causing their female
partners (hanging onto the bucket) to come tumbling after.
Smith recruited 60 pairs of 10-year-old children to participate in his
experiment. Each pair consisted of one boy and one girl. Each pair was
given a bucket and instructed to go up the hill and fetch a pail of water.
Half of the groups were told to “be careful while climbing the hill.” The other
half were simply told to fetch the water with no warning. The pairs were
randomly assigned to a “warning” and “no warning” group. The participants
performed each task separately.
Three observers watched all of the pairs. The observers noted if any of the
children fell down while fetching the water and which child (boy or girl) fell
first.
1. IVs and DVs
2. How are they operationally defined?
3. What type of measures for each variable (N, O, I, R)
18. +
Exercise
List the variables that will be in your study
and place a N, O, I, R to tell me at what level
you plan to measure your variables
21. +
Constructing scales
Determine what you want to measure
Generate a large pool of items
Determine format structure
Have a pool of experts to review it
Administer it to sample
Evaluate items
Optimize scale length
22. +
Scale creation guidelines
Start with twice as many items as you will need
Negative worded items
Every item should reflect the construct
Construct short items (20 words or less)
Avoid emotionally-loaded items
Double-barreled
Avoid using always, never
Avoid double negatives/positives
23. +
Index
Composite measure that summarizes and rank-orders specific
observations and represents some more general dimension
Political Activism
Wrote a letter to a public official
Signed a political petition
Gave money to a political cause
Gave money to a political candidate
Wrote a political letter to the editor
Persuaded someone to change his or her voting plans
25. +
Semantic Differential
Respondents rate their opinions on a linear
scale between two endpoints that have
opposite meanings
Good/Bad, Dirty/Clean, Moral/Immoral
Good 1 2 3 4 5 6 7 Bad
26. +
Measurement error
Error is a combo of random error and measurement
error
Random – error that cannot be predicted or
controlled
Measurement – faulty measurement procedures
32. +
Construct or factorial validity
Operational definition of a variable
Statistical analysis used to see how items correlate with
one another and do not correlate with other items
The degree to which the measurement or manipulation of
the variable accurately reflects the underlying theoretical
construct
Grouped items are called factors (dimensions)
33. +
Criterion validity
The degree to which a measurement device
accurately predicts behavior on a criterion
measure
Predictive
Concurrent
Discriminant Validity
35. +
Test – retest reliability
Take measure two times
Reliability established when the two scores
are very similar
Reliability coefficient – a correlation
coefficient that ranges from 0.00 to 1.00
Highly similar scores are close to 1.00
39. +
Intercoder reliability
Two or more observers/coders judge the
same phenomena
Nominal
Cohen’s kappa and Scott’s pi
Interval and ratio
Coefficient Alpha
Numbers are assigned to things… assign numbers, and you have to create rules on whether and the way numbers are assigned. We care about whether our measures are measuring what they intending to measure
When conceptualize, we specify what every term means. Concepts vary in their level of abstraction, in turn, which affects how easily we can identify indicators to measure them.
Respondents have explicit options from which to choose & it is easier to analyze with stats.
The mathematical precision with which variables can be expressed. The last three progressively become more precise mathematically. You have to decide at which level you will measure each variable. You must be aware whether your variable is at the one of these levels.
The nominal has no mathematical interpretation. A person may be a truck driver or a doctor, but he or she does not represent three units more occupation than the other. Gender is an example of a dichotomous variable. It only have 2 values. Republican is assigned 1 Democrat 2
Every case can only have one attributeEvery case can be classified into one of those categories. “Other” option is often used to ensure that the response are valid.
Permits the researcher not to assess either/or, but rather than greater than or less than. Cultures have to agree that something is less than or greater than which affects the generalizability of the questionnaire
You can add or substract, but because there is no zero…
Core alcohol and drug survey
Divide and multiply… it is more precise, but also have to think of response. Income level should consist of ranges because people do not like to report their income levels
Warning message Falling or number of times falling Gender falling order Jack and Jill went up the hillTo fetch a pail of water.Jack fell down and broke his crown,And Jill came tumbling after Nursery Rhyme
Variable that has more thnan one item
Structurelikert, semantic differential
Sibling communication satisfaction – 15 items, you may have to start with 30 items… I like math.. People will respond to emotions rather than the item. I hate it when my supervisor corrects my mistakesI always communicate competently because no one can honestly say they always communicate compententlyReligious should not have to pay taxes… Religious groups should have to pay taxes would be betterMental Measures Yearbook Buros has a database of mental measures
Adding up scores assigned to individual attributesGive 1 point for each of the actions takenObviously people who areThe key is that you want variance. You want reponses to items to vary.
Latent variable
Attitudinal measurement from Osgood (1952). The most common number of steps is seven. McCroskey and Teven’s (1999) source credibility scale interval level variable that broken down into three dimensions: competence, trustworthiness, caring/goodwill Avoid using jargon such as extraverted/intraverted… talkative/quiet
You learned about measurement, now we want to talk about how to prevent measurement error. The goal is to try to reduce error.Pilot study, seek expert advise, use validated scales,
Researchers say that a measure is valid if it design measure when it meant to measure. Are your measures measure the full essence of the variable? The measure produces stable, consistent reuslts
Mich Alcoholism Screening Test measures 24 questions reflecting the following subscales: recognition of alcohol problems by self & others, legal, social, work problems, help seeking, maritial and family difficulties, liver pathology. Many experts would agree that these dimensions capture the full range of possibilities… the scale is said to have content validity.
Judgment-basedCounting the number of drinks people had consumed in the past week would be a face-valid measure of alcohol consumptionHowever, assessing political competence candidates by how mature their faces look may not be a valid indicator
Theory building at the operationally levelDrop items with low loadings Internal reliability is critical… more so than predictive reliability
Criterion validity is how accurately it predicts a criterion or a well-accepted concept. You calculate the correlation between measures. Measuring the degree to which it predicts another variable. If a particular group rates high on a related variable… and rates high on your variable. It is expected to concurrent validityThe measure of the variable is NOT related to other variables that it theoretically should not be related to.
If a measure if reliable, it is affected less by random error.
Reliability more than .70 or above is means that people are responding to the test in a consistent manner. Ask for a subset of the sample to retake the survey a few weeks later. Reliability is usually calculated where validity is more difficult to assess
Half of the questions in part and the other half in another part. Answers will be 100% in perfect agreement are not ideal, but around 70% is good between halves.
Basically are responding to the same way to related items. Reliabilitycoefficent – a numerical value that tells the percentage of time that measure is reliable 1.00 perfect consistency… 0.00 is no consistency Cronbachs most consistently reported reliability test in the social sciencesMcCroskey… If you use an existing scale you have to report the alpha reliability… and if you create one, you have to report it