Dependencies among indicators of a construct
Locked
Dependencies among indicators of a construct
 
|
Dear all, I am conducting covariance-based SEM using amos. In one of the constructs, I noticed (based on the modification indices) that some indicators/items (five-point Likert type) are dependent on another indicators/items. That is, one item affects the other items. Are dependencies among indicators/items (items from the same construct) not allowed? Thank you. Eins |
Re: Dependencies among indicators of a construct
Administrator
|
I think you need to be more specific!
Which particular MIs are leading you to this conclusion? --
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?" |
Re: Dependencies among indicators of a construct
|
|
Dear David, Suppose X1, X2, X3, X4, and X5 are indicators of F. That is, X1<---F X2<---F X3<---F X4<---F X5<---F The following have MIs of greater than 200: X1-->X3, X3-->X1. This means there are dependencies among X1 and X3. Are these dependencies in the indicators within construct acceptable in SEM? Thank you. E. On Wednesday, December 17, 2014 2:13 PM, David Marso <[hidden email]> wrote: I think you need to be more specific! Which particular MIs are leading you to this conclusion? -- E. Bernardo wrote > Dear all, > I am conducting covariance-based SEM using amos. In one of the constructs, > I noticed (based on the modification indices) that some indicators/items > (five-point Likert type) are dependent on another indicators/items. That > is, one item affects the other items. Are dependencies among > indicators/items (items from the same construct) not allowed? > Thank you.Eins > > ===================== > 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 ----- 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?" -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Dependencies-among-indicators-of-a-construct-tp5728208p5728209.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 |
Re: Dependencies among indicators of a construct
|
Administrator
|
Probably something along the lines of the following:
SET SEED=1615401727. MATRIX. SAVE UNIFORM(1000,5) / OUTFILE * /VARIABLES x1 TO x5. END MATRIX. COMPUTE KSI1=RV.NORMAL(0,1). COMPUTE KSI2=RV.NORMAL(0,1). DO REPEAT x=x1 TO x5. COMPUTE x=.8 * KSI1 + RV.NORMAL(0,.5). END REPEAT. DO REPEAT x=x1 x3. COMPUTE x=x+.5 * KSI2. END REPEAT. i.e. there is something else 'causing' x1 and x3 other than Ksi1 (the putative 'cause' of x1..x5). I suppose you are referring to what is called theta delta in LISREL-ese. Do you have a good theoretical reason for freeing the disturbance term between these two indicators. As a general rule I try to avoid getting involved with discussions of what is acceptable in SEM ;-).
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?" |
Re: Dependencies among indicators of a construct
|
|
In reply to this post by E. Bernardo
Given that you are treating x1-x5 as indicators of a factor, the problem is a (HUGE) residual covariance between x1 and x3. There almost certainly has to be something
odd about x1 and x3 relative to the other items, either item content or in the data itself. Maybe x1 and x3 are near duplicates of each other. Maybe there’s a little recoding error. You should be seeing the problem in the correlation matrix. You can fix the
fit with a residual covariance but you should be prepared justify validity of the scale in the sense of content sampling. Gene Maguin From: SPSSX(r) Discussion [mailto:[hidden email]]
On Behalf Of E. Bernardo Dear David, Suppose X1, X2, X3, X4, and X5 are indicators of F. That is, X1<---F X2<---F X3<---F X4<---F X5<---F The following have MIs of greater than 200: X1-->X3, X3-->X1. This means there are dependencies among X1 and X3. Are these dependencies
in the indicators within construct acceptable in SEM? Thank you. E. On Wednesday, December 17, 2014 2:13 PM, David Marso <[hidden email]> wrote: I think you need to be more specific!
===================== 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 |
Re: Dependencies among indicators of a construct
|
|
In reply to this post by E. Bernardo
In principle, you could introduce causal pathways, but there are other
options. You could introduce and estimate correlations among the error terms, or another latent variable influencing both x1 and x3. That strategy, however, leads to an underidentification problem unless you constrain the coefficients to be equal. David Greenberg, Sociology Department, New York U., On Wed, Dec 17, 2014 at 2:35 AM, E. Bernardo <[hidden email]> wrote: > Dear David, > > Suppose X1, X2, X3, X4, and X5 are indicators of F. That is, > X1<---F > X2<---F > X3<---F > X4<---F > X5<---F > > The following have MIs of greater than 200: X1-->X3, X3-->X1. This means > there are dependencies among X1 and X3. Are these dependencies in the > indicators within construct acceptable in SEM? > > Thank you. > E. > > > On Wednesday, December 17, 2014 2:13 PM, David Marso <[hidden email]> > wrote: > > > I think you need to be more specific! > Which particular MIs are leading you to this conclusion? > -- > > E. Bernardo wrote > >> Dear all, >> I am conducting covariance-based SEM using amos. In one of the constructs, >> I noticed (based on the modification indices) that some indicators/items >> (five-point Likert type) are dependent on another indicators/items. That >> is, one item affects the other items. Are dependencies among >> indicators/items (items from the same construct) not allowed? >> Thank you.Eins > >> >> ===================== >> To manage your subscription to SPSSX-L, send a message to > >> LISTSERV@.UGA > >> (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 > > > > > > ----- > 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?" > -- > View this message in context: > http://spssx-discussion.1045642.n5.nabble.com/Dependencies-among-indicators-of-a-construct-tp5728208p5728209.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 ===================== 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 |
Re: Dependencies among indicators of a construct
|
|
Considering that X1 and X3 are dependent to each other, are they good items? Are they not ambiguous items? |
Re: Dependencies among indicators of a construct
|
Is this a scale you are creating? Or is it a pre-existing scale?
Items for a scale should be correlated but not perfectly. Before getting into SEM, crosstab the items. Are they perfectly correlated? Also run RELIABILITY. Is the iter-item correlation 1.0? Is the squared multiple correlation 1.0? How many cases do you have? Are your respondents representative of the pop the scale was developed for? Is the meaning of the items so close that it makes sense the responses are identically answered. E.g., SD to SA. "I like vanilla ice cream." "I like ice cream flavored with vanilla?" What is the wording of the 5 items? What is the construct they are meant to represent?
Art Kendall
Social Research Consultants |
Re: Dependencies among indicators of a construct
|
|
In reply to this post by E. Bernardo
So, why not publish the items AND their correlation matrix in a post. People will have evaluations. Gene Maguin From: SPSSX(r) Discussion [mailto:[hidden email]]
On Behalf Of E. Bernardo Considering that X1 and X3 are dependent to each other, are they good items? ===================== 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 |
| Free forum by Nabble | Edit this page |