Alpha Question

[Kizzy M Parks]

Hello:
I am writing to solicit your assistance with an ability based survey
pilot.  Before running SEM on my data I need to run reliability on an
ability scale that is scored as right and wrong. I coded the right answer
as 1 and the wrong answer as 0. Using split-half reliability the alpha
score was about .7 however I received an error concerning violating an
assumption due to a negative score.  How may I correct this?

Also, for the second scale there are 4 responses and for some responses 2
items are correct, 1 partially correct and the 4th option is incorrect,
while for others there is one top choice, better choice, acceptable choice
and incorrect.  Any suggestions for coding and running reliability for
this scale?

Thanks!
--
Kizzy M. Parks, Ph.D

Florida Tech

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Art Kendall]

Since the split half reliability estimate was a shorter hand calculation
method to approximate an alpha, I have never used it since I wrote a
FORTRAN program to get an alpha  in 1971.
see interspersed comments.


Kizzy M Parks wrote:
> Hello:
> I am writing to solicit your assistance with an ability based survey
> pilot.  Before running SEM on my data I need to run reliability on an
> ability scale that is scored as right and wrong. I coded the right answer
> as 1 and the wrong answer as 0. Using split-half reliability the alpha
> score was about .7 however I received an error concerning violating an
> assumption due to a negative score.  How may I correct this?
>
>
First I suggest that you check the scoring key. It would be unusual if
an ability item correlated negatively with other items on the same
measure.  If you run RELIABILITY you can see  the correlations among the
items.  The summary of the item correlations will tell you what the
value of the lowest interitem correlation is.  You can help your search
for which items have this score by using <edit> <find> on the output
page.  If either of those items are in many of the negative cells, those
are the best candidate to be mis-keyed.
You can also look at the corrected item-total correlations to look for
negative signs.
> Also, for the second scale there are 4 responses and for some responses 2
> items are correct, 1 partially correct and the 4th option is incorrect,
> while for others there is one top choice, better choice, acceptable choice
> and incorrect.  Any suggestions for coding and running reliability for
> this scale?
>
>
try a DO REPEAT 4 lists 1 with the items, and 1 each for the response
possibilities.  the values
some thing like this untested syntax.  It this first situation the
values to be assigned would be 2, 1, 0
in the second situation the values to be assigned would be 3,2,1,0.

compute score =0.
DO REPEAT item = item1 to item10
  /For_a = 0 3 2 0 3 1 2 3 0 0
  /For_b = 1 2 3 1 2 3 1 2 3 2
  /For_c = 3 2 1 0 0 1 2 3 1 1
  /For_d = 3 1 0 2 3 1 0 2 3 1.
do   if item eq 'a'.
compute score = score + For_a.
else if item eq 'b'.
compute score = score + For_b.
else if item eq 'c'.
compute score = score + For_c.
else if item eq 'd'.
compute score = score + For_d.
else,
print / 'oops bad response value ID= ' ID item.
end if.

Art Kendall
Social Research Consultants

> Thanks!
> --
> Kizzy M. Parks, Ph.D
>
> Florida Tech
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>
>

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD