Brief Summary of Chapter 5
Chapter 5 primarily focuses on construct validity.
One of the hardest things to do in psychology is to establish construct validity. For example, how do you show that you are really measuring a mental construct, such as love?
You may start by getting an operational definition--a recipe that you will follow to get a score for each participant. By using an operational definition, you hope that your measure will be objective (free from observer bias). But it might not be! And, even if it's objective, it might not be valid.
To make a case for your measure's construct validity, you may try to establish that your measure:
Back to Chapter 5 Menu
.
.
.
.
.
.
.
.
.
.
.
If observer bias seems to be a problem, consider changing your measure (see table
5-1 on page 153 to learn how) or making observers "blind. "
Go back
.
.
.
.
.
.
.
.
.
Even if your measure is free from observer bias, it may not be free of subject bias.
For example, a multiple-choice personality measure could be objectively scored, but participants could fake their responses.
Many strategies to avoid subject bias are described Table 5-2 (p. 157) . Another approach to the subject bias problem is to make participants blind.
Bias is the most serious type of error a measure can have. It can cause researchers to get the results they expect or participants to produce the results they think the researcher wants. If researchers are just going to bias the results so that they get the results they want, there's no point in doing the study! However, bias is not the only type of error a measure can have.
Go back
.
.
.
.
.
Another type of error that can contaminate a measure's scores is random error.
Random error is unsystematic. It won't consistently push scores up or down. Instead, in an unpredictable way,
it pushes some scores up and others down. We can estimate the extent to which a measure is influenced by random error by
calculating reliability coefficients. Perhaps the most common strategy for getting a sense of
the full extent of random error is to calculate a test-retest reliability coefficient.
If test-retest reliability is low, we are concerned. We worry that if the measure is so vulnerable to random error,
it might also be vulnerable to some other errors. We are also worried because we know that when the measure is
being affected by random error, it is NOT being affected by our construct.
If the measure is not reliable, then we may
- Ditch the measure
- Try to find out the source of the random error and reduce it. To learn how, see Figure 5-6 (page 175).
If the reliability is high, we know the measure is measuring SOMETHING consistently. But what?
Go back
.
.
.
.
.
.
.
content validity: the measure has items that cover all of the relevant
dimensions of what you are trying to measure and there are enough items for each dimension.
Go back
.
.
.
.
.
.
.
.
.
.
internal consistency: all the items seem to be measuring the same thing.
If participants who strongly agreed with one item strongly agreed with most of the other items, whereas participants who
strongly disagreed with one item strongly disagreed with most of the other items, that pattern would be evidence of the
measure's internal consistency. One way of thinking of internal consistency is to view each item as a judge
(as in the Olympics). If all the "judges" agree, we are more confident of the total score. However, if some "judges" (items)
claim that the participant is very shy whereas other "judges" claim that the participant is very outgoing, we would have doubts
about the total score. The judges inconsistency would bother us.
Go back
.
.
.
.
.
convergent validity: the measure correlates with other indicators of the construct.
If you have a test of IQ, participants who score high on your IQ test should score higher on other tests of IQ than participants who
scored low on your test.
Go back
.
.
.
.
.
.
.
.
.
.
discriminant validity: Trying to show that you aren't measuring the wrong construct
by showing that your measure doesn't correlate too highly with measures of these "wrong" constructs. For example, suppose you want to show
that your measure of intelligence isn't just a measure of social desirability. You would hope that, relative to people who get average
or low scores on your IQ tests, that people who score high on your IQ test should not also get higher scores on the social desirability test.
Go back
Back to Chapter 5 Menu