 |
|
| Science Forum Index » Statistics - Education Forum » Alphas for repeated administrations of an instrument... |
|
Page 1 of 1 |
|
| Author |
Message |
| gingerinberkeley... |
 Posted: Wed Aug 05, 2009 2:15 pm |
|
|
|
Guest
|
Hello,
Sorry, I'm not a statistics person but have a question I hope someone
will be able to assist me with.
We have conducted research using an psychological instrument that was
administered at baseline, 6-, 12- and 18-month follow-ups (4 time
points; n=350). The reliability of this instrument has been
established in the literature (using Cronbach's alphas) but only using
baseline data. No one has ever used this instrument longitudinally.
Is it necessary to establish this instrument as reliable over time
(Cronbach's alphas for each time period?) ?
Thanks in advance for any advice. |
|
|
| Back to top |
|
|
|
| Ray Koopman... |
| Posted: Wed Aug 05, 2009 6:41 pm |
|
|
|
Guest
|
On Aug 5, 5:15 pm, gingerinberkeley <gingerinberke... at (no spam) gmail.com>
wrote:
[quote:ecfd0ee59f]Hello,
Sorry, I'm not a statistics person but have a question I hope someone
will be able to assist me with.
We have conducted research using an psychological instrument that was
administered at baseline, 6-, 12- and 18-month follow-ups (4 time
points; n=350). The reliability of this instrument has been
established in the literature (using Cronbach's alphas) but only using
baseline data. No one has ever used this instrument longitudinally.
Is it necessary to establish this instrument as reliable over time
(Cronbach's alphas for each time period?) ?
Thanks in advance for any advice.
[/quote:ecfd0ee59f]
Several comments:
1. The validity of a sample alpha depends strongly on the assumption
that the errors of measurement in the items correlate only negligibly.
For many instruments, it seems more likely that those errors will be
positively correlated, which will lead to overestimation of alpha.
(This is a problem that psychometricians have mostly ignored.)
2. Even when it is estimated accurately, alpha is not a reliability,
but a lower bound for the reliability. An instrument whose items were
all perfectly reliable but mutually uncorrelated would have an alpha
of zero but would still be perfectly reliable.
3. For longitudinal work, that considers changes in individuals over
time, you should worry about the standard error of measurement (SEM),
not reliability. Reliability is an artifact of the heterogeneity of
the group in which it was assessed and does not apply to individuals'
scores.
4. Yes, it would help to have information about the temporal stability
of individuals' scores under "natural" conditions, such as might be
obtained from some sort of control group that was measured at the same
time intervals. In the absence of such information, you may find it
difficult to draw strong conclusions, even for individuals whose
changes are substantially larger than the SEM. |
|
|
| Back to top |
|
|
|
|
|
All times are GMT - 5 Hours
The time now is Mon Dec 07, 2009 12:17 pm
|
|