Lecture 2:  Constructs and Operational Definitions

Outline
Operational Definitions
     definitions
     what you gain
    scales of measurement
     operationalize that construct
     Difficulties in scoring/coding
Questions
    Why ask questions
    Overall design
    Types of responses
    Wording instructions, questions, and responses
    How many questions
    Pitfalls
 

Definitions
Constructs:  theoretical concept; a construction of a cause; can never be measured directly; have no evidence that it is a real thing, even though we tend to act as though it were
Op Defs:  concrete representation of a construct; a way of counting & measuring behaviors

Even constructs that seem straightforward are sometimes tricky:  Mother

Why is this important in Psychology?  Because most of our variables are constructs and we have to operationalize them in some fashion for easy and consistent measurement.  We have no way of knowing these are real things.

Memory, Personality, Hunger, Emotion, Intelligence, Learning, Friendship, Love

Op Defs can come in many forms:
 behavior counts
 physiological measures
 self reports
 scales
 observer ratings

Op Defs need to be
 Precise
 Cover full range
 Do 1 thing at a time
 Not confuse constructs

What you gain
 allows count
 allows reliability (others can count same and see if we agree),
 allows replication (others know what I did),
 allows construct validity (can determine if it measures what it claims to)

Scales of measurement
 I should say a word about scales of measurement at this point.
  Nominal -- just grouping
  Ordinal -- ordered but not consistent differences
  Interval -- consistent differences without zero
  Ratio -- consistent differences with zero
All stat books talk about this issue -- so make sure that you are familiar with it.  You generally want at least interval variables in you DVs so we can conduct the stats we will cover

Operationalize that construct

Hunger, love, extroversion

Difficulties in scoring/coding

Example 1:  Memory for a word list

Had Ss listen to and write down a word list
What counts as a correct answer?

 List:  plan, tops, fin, hat, runt, kid, sap, lisp, ram, cup

 Answers:   plant, tops, fin, hat, lips, cup

Example 2: Coding childrenís memory content

Tape record mother-child conversations
Classify types of activities:  Actions, Verbalization, Cognitive/emotional attributions

 I rode my bike
 She went go away
 I was like I donít know
 She seemed upset

Questionnaires

Why ask questions
Another way of making OpDefs
 ask about demographics
 ask about opinions and intentions
 ask about behaviors
Very useful for things that are hard or tricky to measure (such as opinions, or real behaviors like reproductive rates)

Overall design
when asking questions, particularly in a questionnaire of some sort, you should start at the broad level

 stay on topic
1.  only include questions about the specific topic
 make sure that you intend to use a question
2.  be narrow
3.  keep it short

 ordering (do it to make it clear and easy)
put clear explanation and confidentiality statement at the front
put boring demographic questions at the end
group questions by topic -- donít switch topics often
give information when you switch topics
if you ask threatening questions, put them after related non-threatening questions
watch for biasing in order (rating of President; Puget Power building lines)
**exception is if you are trying to hide the goal of your questionnaire

 look
make it clean
make the questions in very similar formats
put the answers below the questions, maybe indented, maybe all CAPS
use spacing
 

Types of responses
you can distinguish among questionnaires by the type of responses

 open-ended
let the subjects answer as they please
 good -- because you get the subjectís view rather than your own
 bad -- because you then have to group the responses in some fashion

 forced choice or restricted item
provide subjects with alternatives
if you can -- put the responses in order (never, 1-2 time, 3-5 times, more than 5)
 good -- easy to score
 bad -- limit Sís and you may miss some possibilities

 partially open-ended
provides an Ďotherí category in restricted item format

 scaled
allow subjects to choose from a range of options
number of points on the scale:  from 3 to 100
 rec. at least 7 (with 5 if people donít use ends you have little variability)
 probably no more than 10

number of anchors
 endpoints?  all points?  endpoints and middle?
 only vary one dimension (cut defense to raise social  <--> raise defense to save us from communism) (save jobs <--> save the owl)
odd or even?
 odd leaves a true middle
 even forces a side

***Note that there are lots of variations on these themes in terms of responses***

Wording instructions, questions, and responses
 clear
donít use big words or infrequently used words
ask about one thing at a time
donít use no negatives

 non-biased
be careful to not force a socially acceptable response in your wording
 (Pat Swindall survey:  Are you in favor of supporting the freedom fighters in Nicaragua in their battle to throw out the communist dictators?)

How many questions
If you are doing this to answer one specific question -- ask only that one
 ---  who will you vote for?
If you are trying to measure a construct, ask many.
 why more than one
just one is too random (misread, individual differences),
asking more gives you a better idea
 same thing as stepping on the scale three times and averaging
ask slightly different OpDefs of the same Construct
 how related (Likert scale)
make sure the questions are related
Make sure that you switch direction on some
 additive??
generally you want your answers to be on the same scale so that, if they are only measuring one construct, you can add them together

Pitfalls
presentation problems -- subjects donít tell the truth, subjects donít know, subjects donít remember, subjects donít care

measurement problems
  -- you arenít measuring just one thing (construct valididity)
  -- not a consistent measurement (reliability)

Variants on correlations to solve measurement problems
Reliability:  split-half or test-retest
Validity:  Correlate to other measures