june 14 lecture

9 views2 pages
9 Apr 2012
of 2
Next week midterm!
-writing in AA112
-held during class time
-first 5 chapters
-this lecture included
>Operational Definitions
Reliability: the proportion of truth in a measure; reliability is looked at over a large number
of subjects – leads to a variability in scores
-not looking at mean scores but how variance differs in scores
-multiple choice questions on reliability, not short answer
-the proportion of true score in each measure
-measurement error varies each time
-true score (we don’t know) is the same, especially over a short period of time
-*know term true score
-variance of true score (T)/ Variance of measure (X)
-can we calculate actual (true) reliability of a measure? No, can’t calculate variance of true
scores because won’t know what they are
-can estimate it
-can assume our measurement scores share a true score
-it is the measurement error that differs
-measurement scores are related to each other to the degree that they share true scores
Estimate’s of reliability – reliability coefficients (pearson’s coefficients)
-variance in true scores/ variance in x
-x = true score + measurement error
=variance of true score/ variance of true score + variance of measurement error
i.e. two scores on a test; first time you score 85, second time you only score 80. x1 = 85,
x2=80 – these are sample measures. Factors (i.e. stress, sleep, studying time, etc) each
measurement score consists of a true score, so you’re actual ability in math + measurement
So T + e1
Second time you wrote same test, T + e2, so maybe this time you got less sleep.
These two scores (T’s) are the same; but we don’;t know those scores. E1 and e2 are
random, so they may differ each time. When estimating reliability, we can assume that the
true scores in these measures are the same. It is the error that differs. So as an estimate of
reliability, the best we can do is to look at covariance between two scores, and we can
calculate that correlation, and that is a reasonable estimate of reliability of that measure. So
ratio between two true scores is consistent, it is error that is different.
-the higher the correlation, the higher the estimated reliability
Internal Consistency Reliability
1. Test-retest reliability – vulnerable to artificiality because person is doing same test, so
subject may want to please experimenter and will attempt to write test differently;
Maturation – changes that might happen; Parallel forms – same construct but different