# reliability analysis

10 messages
Open this post in threaded view
|
Report Content as Inappropriate

## reliability analysis

 While performing reliability analysis, I am getting the message as.. The determinant of the covariance matrix is zero or approximately zero. Statistics based on its inverse matrix cannot be computed and they are displayed as system missing values. No. of observations are 320 and Cronbach's Alpha is .581. ( Cronbach's Alpha Based on Standardized Items is .565.) And, none of the correlation is >.9. Please let me know what is wrong with data ? thanx vini
Open this post in threaded view
|
Report Content as Inappropriate

## Re: reliability analysis

 Are you sure that you did not enter a variable name twice? The multiple correlation for some variable is around 1.00. What variable(s) has/have a high corrected item-total correlation?  look there How many items do you have? If you only have a small number of variables:  specify the variable list in full (no TO). cut and paste the scale command as many times as you have suspect variables, cutting out one variable that has a high item-total correlation from each. IF worst comes to worst do as many scale specifications as there are variables omitting one at a time. if these things do not work, re-post to the list and someone can explain how to create a purely random Y variable that you can use REGRESSION and its diagnostics to track it down.  But first try the easier things. ```Art Kendall Social Research Consultants``` On 7/22/2012 1:47 AM, vini wrote: ```While performing reliability analysis, I am getting the message as.. The determinant of the covariance matrix is zero or approximately zero. Statistics based on its inverse matrix cannot be computed and they are displayed as system missing values. No. of observations are 320 and Cronbach's Alpha is .581. ( Cronbach's Alpha Based on Standardized Items is .565.) And, none of the correlation is >.9. Please let me know what is wrong with data ? thanx vini -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/reliability-analysis-tp5714364.html Sent from the SPSSX Discussion mailing list archive at Nabble.com. ===================== 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 Art Kendall Social Research Consultants
Open this post in threaded view
|
Report Content as Inappropriate

## Re: reliability analysis

 In reply to this post by vini Thanks for your reply. None of the variables entered twice. The multiple correlation ranges from -.068 to .638. The highest corrected item-total correlation is .533. Is this suspect variable? Not getting "cut and paste scale command". However, omitting one variable at a time with high multiple item total correlation does not help. there are 13 variables and all are dichotomous bearing values aither 1 or 2.                         Summary Item Statistics                             Mean Minimum Maximum Range Maximum / Minimum Variance                                                                                                                       N   of Items Item Means      1.583 1.013 1.994 .981 1.969    .141                    13 Item Variances      .113              .006 .246 .240 39.491    .010                    13 Inter-Item Covariances  .011             -.010 .138 .148 -14.194    .001                    13 Inter-Item Correlations   .090 -.209 .638 .847 -3.057    .029                    13 Hope this table would help you to understand the problem better, Thanks Vini
Open this post in threaded view
|
Report Content as Inappropriate

## Re: reliability analysis

 Vini, I didn't see your initial message so I'm sure there is something I don't know (and nabble is less useful than a listserv since prior message aren't included but anyway ...) BUT aren't you a little surprised to see at least one negative correlation between scale variables? Gene Maguin -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of vini Sent: Wednesday, July 25, 2012 1:45 AM To: [hidden email] Subject: Re: reliability analysis Thanks for your reply. None of the variables entered twice. The multiple correlation ranges from -.068 to .638. The highest corrected item-total correlation is .533. Is this suspect variable? Not getting "cut and paste scale command". However, omitting one variable at a time with high multiple item total correlation does not help. there are 13 variables and all are dichotomous bearing values aither 1 or 2.                         Summary Item Statistics                             Mean        Minimum Maximum Range   Maximum / Minimum       Variance N   of Items Item Means            1.583     1.013   1.994   .981    1.969       .141     13 Item Variances        .113       .006      .246    .240    39.491      .010   13 Inter-Item Covariances  .011   -.010      .138    .148    -14.194     .001    13 Inter-Item Correlations   .090  -.209   .638    .847    -3.057      .029   13 Hope this table would help you to understand the problem better, Thanks Vini -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/reliability-analysis-tp5714364p5714438.htmlSent from the SPSSX Discussion mailing list archive at Nabble.com. ===================== 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
Open this post in threaded view
|
Report Content as Inappropriate

## Re: reliability analysis

Open this post in threaded view
|
Report Content as Inappropriate

## Re: reliability analysis

 In reply to this post by vini It seems very unusual that you would have a single interitem correlation of .638 and a  squared multiple correlation of .638. How did you get the multiple correlation on runs that on which the matrix could not be inverted? An item-total correlation is not a multiple correlation . It is a bivariate correlation between an item and and a single variable which is the sum of the other items. With regard to the negative correlation  of -.209, that is also something to look into. Are you sure that the items are not balanced or at least varied for wording direction? Does that item have negative interitem correlations with several variables? A major part of the alpha is the mean of all the inter item correlations so negative interitem correlations could be a reason you alpha is so low? What does the output show as the number of cases? Are there items that if they were deleted the alpha would be higher? What are the  variable labels for the 13 items? Do you mean that you did 13 reliability runs each on 12 variables? ```Art Kendall Social Research Consultants``` On 7/25/2012 1:45 AM, vini wrote: ```Thanks for your reply. None of the variables entered twice. The multiple correlation ranges from -.068 to .638. The highest corrected item-total correlation is .533. Is this suspect variable? Not getting "cut and paste scale command". However, omitting one variable at a time with high multiple item total correlation does not help. there are 13 variables and all are dichotomous bearing values aither 1 or 2. Summary Item Statistics Mean Minimum Maximum Range Maximum / Minimum Variance N of Items Item Means 1.583 1.013 1.994 .981 1.969 .141 13 Item Variances .113 .006 .246 .240 39.491 .010 13 Inter-Item Covariances .011 -.010 .138 .148 -14.194 .001 13 Inter-Item Correlations .090 -.209 .638 .847 -3.057 .029 13 Hope this table would help you to understand the problem better, Thanks Vini -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/reliability-analysis-tp5714364p5714438.html Sent from the SPSSX Discussion mailing list archive at Nabble.com. ===================== 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 Art Kendall Social Research Consultants
Open this post in threaded view
|
Report Content as Inappropriate

## Re: reliability analysis

 In reply to this post by vini  - I am reposting this because the List never sent meconfirmation, and I don't see the post at Nabble.**********I'm trying to figure these summary statistics. I see a dichotomous variable, (1 2), with a mean of 1.994. That is not hardly useful for inter-item correlations unless other means are in the same range.  But the range includes the other extreme, too -- 1.013.  Dichotomies cannot correlate well (which is the basis of alpha) when their skews differ by much.  Those two variables  might be important in some sense but they cannot contribute usefully to the alpha: Try it without. I wonder if they are responsible for screwing up some algorithm.  "The multiple correlation ranges from -.068 ...." THERE is a variable to omit.  Multiple correlations are restricted to non-negative values, so the program has definitely screwed up, if that is really what it says.  Or, you (maybe) are supposed to ignore those statistics because of the warning about the determinant. It might be time to try this on another computer, because if SPSS is not corrupted, the procedure may have an error, despite the testing it has seen.  I think it should not publish a multiple correlation if it cannot give a valid one. -- Rich Ulrich   > Date: Tue, 24 Jul 2012 22:45:05 -0700 > From: [hidden email] > Subject: Re: reliability analysis > To: [hidden email] > > Thanks for your reply. > None of the variables entered twice. The multiple correlation ranges from > -.068 to .638. The highest corrected item-total correlation is .533. Is this > suspect variable? > Not getting "cut and paste scale command". However, omitting one variable at > a time with high multiple item total correlation does not help. > there are 13 variables and all are dichotomous bearing values aither 1 or 2. > > Summary Item Statistics > Mean Minimum Maximum Range Maximum / Minimum Variance N of Items                      mean  min    max  range  ratio var  N_i > Item Means 1.583 1.013 1.994 .981 1.969 .141 13 > Item Variances .113 .006 .246 .240 39.491 .010 13 > Inter-Item Covariances .011 -.010 .138 .148 -14.194 .001 13 > Inter-Item Correlations .090 -.209 .638 .847 -3.057 .029 13 >  ...
Open this post in threaded view
|
Report Content as Inappropriate

## Re: reliability analysis

 In reply to this post by vini Sorry ! my mistake. It was not multiple correlation but bivariate as rightly said by Kendall. In fact, I am not getting values for squared multiple correlations. As far as the item with a negative correlation of -.209 is concerned, this item has negative correlation with 3 other items too. There are no other items that can make alpha higher if deleted ( only marginal improvement from .581 to .588) Yes I did 13 reliaility runs each on 12 variables and warning message remains there. Even omitting 2 variables (with lowest and highest mean) together and running reliability does not stop this warning message, however improves alpha value a bit. Please suggest some way out ! Thanx Vini
Open this post in threaded view
|
Report Content as Inappropriate

## Re: reliability analysis

 The warning means that one or more of the variables is perfectly predictable from some combination of the others!!! Find that variable and lose it!  Beyond that it is probably NECESSARY or the list to see the ACTUAL correlation matrix or the raw data to do much.  You really need to do your research and look up the meaning of determinant, singular matrix and matrix inverse! vini wrote Sorry ! my mistake. It was not multiple correlation but bivariate as rightly said by Kendall. In fact, I am not getting values for squared multiple correlations. As far as the item with a negative correlation of -.209 is concerned, this item has negative correlation with 3 other items too. There are no other items that can make alpha higher if deleted ( only marginal improvement from .581 to .588) Yes I did 13 reliaility runs each on 12 variables and warning message remains there. Even omitting 2 variables (with lowest and highest mean) together and running reliability does not stop this warning message, however improves alpha value a bit. Please suggest some way out ! Thanx Vini Please reply to the list and not to my personal email. Those desiring my consulting or training services please feel free to email me. --- "Nolite dare sanctum canibus neque mittatis margaritas vestras ante porcos ne forte conculcent eas pedibus suis." Cum es damnatorum possederunt porcos iens ut salire off sanguinum cliff in abyssum?"
 In reply to this post by vini There are several things that appear unusual in you data. You have at least 2 items with almost no variance. One with almost all 1's and one with all 2's. Where did the set of 13 items come from?  Is this a previously used scale? If it is a previously used scale did the original authors say some items need to be reflected? What is being measured? I.e., what is the underlying construct? What are the value labels for the items? 1 means ______. 2 means ______. What are the variable labels? Which item has the negative correlations? If you send me an anonymized system file with the variables view completed I'll take a look at it. Who are the respondents? How were they chosen? Do you have any items that have some missing values? If you send me an anonymized system file with the variables view completed I'll take a look at it. ```Art Kendall Social Research Consultants``` On 7/26/2012 1:13 AM, vini wrote: ```Sorry ! my mistake. It was not multiple correlation but bivariate as rightly said by Kendall. In fact, I am not getting values for squared multiple correlations. As far as the item with a negative correlation of -.209 is concerned, this item has negative correlation with 3 other items too. There are no other items that can make alpha higher if deleted ( only marginal improvement from .581 to .588) Yes I did 13 reliaility runs each on 12 variables and warning message remains there. Even omitting 2 variables (with lowest and highest mean) together and running reliability does not stop this warning message, however improves alpha value a bit. Please suggest some way out ! Thanx Vini -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/reliability-analysis-tp5714364p5714462.html Sent from the SPSSX Discussion mailing list archive at Nabble.com. ===================== 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 Art Kendall Social Research Consultants