Interpreting interactions with covariate
Locked
Interpreting interactions with covariate
 
|
Hi,
I have several dependent variables for which I have conducted 2x2x2x2 ANCOVAs with the same IVs and covariate using the GLM command. In several of these ANCOVAs, there are significant 4-way interactions that I am trying to interpret. It is my understanding that I should try to tease apart the effects by splitting the file by one of my IVs and working my way down the lower order interactions to see where the effects are. However, I am lost with regard to the covariate. Should I be trying to see the effects solely at the the mean of the covariate? How can I describe the patterns when they are likely changing at different values of the covariate? Additionally, is there a reason why SPSS does not test for interactions between the covariate and between-subject IVs? And how do I interpret an interaction between my within-subject IVs and my covariate? I have thought about plotting my covariate along the horizontal axis with different lines for the levels of an IV, but this is not as simple as seeing that there are different slopes. Any advice would be so greatly appreciated! Jillian ===================== 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: Interpreting interactions with covariate
Administrator
|
Hello Jillian. One option would be to use LMATRIX and MMATRIX sub-commands,
as shown by Howell & Lacroix (2012). http://www.tqmp.org/RegularArticles/vol08-1/p001/p001.pdf But personally, I find the requirement of generating all those coefficients off-putting, and a likely source of error. Interaction contrasts like this are much easier to do in Stata, IMO. E.g., the code might look something like this: // ANCOVA model with x as continuous covariate anova y c.x a##b##c##d // a*b*c interaction at each level of d contrast a#b#c@d, effects nowald // a*b interaction for each c*d combination contrast a#b@c#d, effects nowald // Effect of a for each b*c*d combination contrast a@b#c#d, effects nowald // Etc. Jon, if you're looking for extension command ideas, something that performs interaction contrasts like this more easily would be a good one, IMO. HTH. Jillian Ma wrote > Hi, > > I have several dependent variables for which I have conducted 2x2x2x2 > ANCOVAs with the same IVs and covariate using the GLM command. In several > of these ANCOVAs, there are significant 4-way interactions that I am > trying to interpret. It is my understanding that I should try to tease > apart the effects by splitting the file by one of my IVs and working my > way down the lower order interactions to see where the effects are. > > However, I am lost with regard to the covariate. Should I be trying to see > the effects solely at the the mean of the covariate? How can I describe > the patterns when they are likely changing at different values of the > covariate? > > Additionally, is there a reason why SPSS does not test for interactions > between the covariate and between-subject IVs? And how do I interpret an > interaction between my within-subject IVs and my covariate? I have thought > about plotting my covariate along the horizontal axis with different lines > for the levels of an IV, but this is not as simple as seeing that there > are different slopes. > > Any advice would be so greatly appreciated! > > Jillian > > ===================== > 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 ----- -- Bruce Weaver [hidden email] http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." NOTE: My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. -- Sent from: http://spssx-discussion.1045642.n5.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
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
Re: Interpreting interactions with covariate
|
|
Thanks, Bruce. This resource seems very helpful to try to tackle the LMATRIX and MMATRIX subcommands. I have still not figured out what to do now with a covariate. I believe I cannot simply look at the interaction at all levels of the covariate because my covariate is a continuous variable. Please let me know if you have any clues on what to do with this covariate. On Fri, Jul 27, 2018 at 9:23 AM, Bruce Weaver <[hidden email]> wrote: Hello Jillian. One option would be to use LMATRIX and MMATRIX sub-commands, |
|
|
In reply to this post by Bruce Weaver
Well, the SPSSINC CREATE DUMMIES (Transform > Create Dummy Variables) will create dummy variables and 1, 2, and 3-way interaction variables. It can also generate macro names for each set. But it doesn't directly generate the contrast specifications. On Fri, Jul 27, 2018 at 7:24 AM Bruce Weaver <[hidden email]> wrote: Hello Jillian. One option would be to use LMATRIX and MMATRIX sub-commands, |
Re: Interpreting interactions with covariate
|
|
In reply to this post by Jillian Ma
hi Jillian, using
/PRINT PARAMETER
subcommand will give you parameter estimates which can help you interpret the impact of the covariate. You may also check the following link for more info on how to interpret them.
hth,
bozena
From: SPSSX(r) Discussion [[hidden email]] on behalf of Jillian Ma [[hidden email]]
Sent: Friday, July 27, 2018 9:48 AM Thanks, Bruce. This resource seems very helpful to try to tackle the LMATRIX and MMATRIX subcommands. I have still not figured out what to do now with a covariate. I believe I cannot simply look at the interaction at all levels of the covariate
because my covariate is a continuous variable. Please let me know if you have any clues on what to do with this covariate.
On Fri, Jul 27, 2018 at 9:23 AM, Bruce Weaver
<[hidden email]> wrote:
Hello Jillian. One option would be to use LMATRIX and MMATRIX sub-commands, |
Re: Interpreting interactions with covariate
|
Administrator
|
In reply to this post by Jillian Ma
Does your model include any interactions involving the covariate? If not,
then the nature of the interactions involving your 4 factor variables does not depend on the value of the covariate, and should be able to proceed as shown in the Howell & Lacroix article, /I think/. Jillian Ma wrote > Thanks, Bruce. This resource seems very helpful to try to tackle the > LMATRIX and MMATRIX subcommands. I have still not figured out what to do > now with a covariate. I believe I cannot simply look at the interaction at > all levels of the covariate because my covariate is a continuous variable. > Please let me know if you have any clues on what to do with this > covariate. > > On Fri, Jul 27, 2018 at 9:23 AM, Bruce Weaver < > bruce.weaver@ > > > wrote: > >> Hello Jillian. One option would be to use LMATRIX and MMATRIX >> sub-commands, >> as shown by Howell & Lacroix (2012). >> >> http://www.tqmp.org/RegularArticles/vol08-1/p001/p001.pdf >> >> But personally, I find the requirement of generating all those >> coefficients >> off-putting, and a likely source of error. Interaction contrasts like >> this >> are much easier to do in Stata, IMO. E.g., the code might look something >> like this: >> >> // ANCOVA model with x as continuous covariate >> anova y c.x a##b##c##d >> // a*b*c interaction at each level of d >> contrast a#b#c@d, effects nowald >> // a*b interaction for each c*d combination >> contrast a#b@c#d, effects nowald >> // Effect of a for each b*c*d combination >> contrast a@b#c#d, effects nowald >> // Etc. >> >> >> Jon, if you're looking for extension command ideas, something that >> performs >> interaction contrasts like this more easily would be a good one, IMO. >> >> HTH. >> >> >> Jillian Ma wrote >> > Hi, >> > >> > I have several dependent variables for which I have conducted 2x2x2x2 >> > ANCOVAs with the same IVs and covariate using the GLM command. In >> several >> > of these ANCOVAs, there are significant 4-way interactions that I am >> > trying to interpret. It is my understanding that I should try to tease >> > apart the effects by splitting the file by one of my IVs and working my >> > way down the lower order interactions to see where the effects are. >> > >> > However, I am lost with regard to the covariate. Should I be trying to >> see >> > the effects solely at the the mean of the covariate? How can I describe >> > the patterns when they are likely changing at different values of the >> > covariate? >> > >> > Additionally, is there a reason why SPSS does not test for interactions >> > between the covariate and between-subject IVs? And how do I interpret >> an >> > interaction between my within-subject IVs and my covariate? I have >> thought >> > about plotting my covariate along the horizontal axis with different >> lines >> > for the levels of an IV, but this is not as simple as seeing that there >> > are different slopes. >> > >> > Any advice would be so greatly appreciated! >> > >> > Jillian >> > >> > ===================== >> > 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 >> >> >> >> >> >> ----- >> -- >> Bruce Weaver >> > bweaver@ >> http://sites.google.com/a/lakeheadu.ca/bweaver/ >> >> "When all else fails, RTFM." >> >> NOTE: My Hotmail account is not monitored regularly. >> To send me an e-mail, please use the address shown above. >> >> -- >> Sent from: http://spssx-discussion.1045642.n5.nabble.com/ >> >> ===================== >> 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 >> > > ===================== > 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 ----- -- Bruce Weaver [hidden email] http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." NOTE: My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. -- Sent from: http://spssx-discussion.1045642.n5.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
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
Re: Interpreting interactions with covariate
|
|
That makes sense to me. The model does include a significant interaction between the two within-subject IVs and the covariate. I might be able to explore this with some plotting. I do wonder why by default SPSS does not tell me about any interactions between the between-subject IVs and the covariate. On Fri, Jul 27, 2018 at 2:46 PM, Bruce Weaver <[hidden email]> wrote: Does your model include any interactions involving the covariate? If not, |
Re: Interpreting interactions with covariate
|
|
Would you please post your model statement? There’s suddenly appeared two phrases (“significant interaction” and “within-subject”). Given what you’d posted earlier
I pictured something like this: GLM y by x1 x2 x3 x4 with cov.
Four between factors with a covariate, all interactions among factors, no factor by covariate interactions. Gene Maguin From: SPSSX(r) Discussion <[hidden email]>
On Behalf Of Jillian Ma That makes sense to me. The model does include a significant interaction between the two within-subject IVs and the covariate. I might be able to explore this with some plotting. I do wonder why by default SPSS does not tell me about any
interactions between the between-subject IVs and the covariate. On Fri, Jul 27, 2018 at 2:46 PM, Bruce Weaver <[hidden email]> wrote:
===================== 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: Interpreting interactions with covariate
|
In reply to this post by Jillian Ma
About the meaning of parameters and types of contrasts and
corresponding coding - you may want to read
https://stats.stackexchange.com/a/221868/3277 ===================== 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: Interpreting interactions with covariate
|
|
In reply to this post by Maguin, Eugene
GLM a1b1 a1b2 a2b1 a2b2 by c d with cov. 2 within factors, 2 between factors and a covariate. Looking at all possible interactions. Very appreciative of all the help here! On Fri, Jul 27, 2018 at 3:07 PM, Maguin, Eugene <[hidden email]> wrote:
|
Re: Interpreting interactions with covariate
|
|
In reply to this post by Jillian Ma
I think SPSS does not, by default, say anything about interactions between the IVs and the
covariate because that reeks of data-dredging.
Looking at interactions, unless it is so that you assert that there weren't any, also tends to smell of data-dredging, especially when you already have 4 main effects. Also, effects of 3rd
order and above tend to be hard to interpret. Moreover, the higher the order, the more likely that the effect is not replicable - unless it contains a common artifact.
Bad scaling can create interaction-artifacts, so I'd check for outliers or other suggestions that transformation is needed. (Using Ranks is bad for multi-variable analyses because that creates interactions, by artifact.)
Having a mixture of two populations that are too dissimilar can also give you higher order interactions. In that case, doing separate analyses is often the cure.
-- Rich Ulrich
From: SPSSX(r) Discussion <[hidden email]> on behalf of Jillian Ma <[hidden email]>
Sent: Friday, July 27, 2018 3:01 PM To: [hidden email] Subject: Re: Interpreting interactions with covariate That makes sense to me. The model does include a significant interaction between the two within-subject IVs and the covariate. I might be able to explore this with some plotting. I do wonder why by default SPSS does not tell me about any interactions
between the between-subject IVs and the covariate.
[snip, rest]
|
Re: Interpreting interactions with covariate
|
Administrator
|
In reply to this post by Jillian Ma
To give the info Gene is asking for, you need to include the complete syntax
for your command. Specifically, you need to show the WSFACTOR, WSDESIGN and DESIGN sub-commands. That should clarify what you mean when you say "all possible interactions". (I think what you mean by that is all possible interactions involving the factor variables, but no interactions involving the covariate.) The command syntax should be echoed in your output window, so you could copy from there. Alternatively, exit the GUI dialog via the PASTE button to generate the syntax. HTH. Jillian Ma wrote > GLM a1b1 a1b2 a2b1 a2b2 by c d with cov. > 2 within factors, 2 between factors and a covariate. Looking at all > possible interactions. > > Very appreciative of all the help here! > > On Fri, Jul 27, 2018 at 3:07 PM, Maguin, Eugene < > emaguin@ > > wrote: > >> Would you please post your model statement? There’s suddenly appeared >> two >> phrases (“significant interaction” and “within-subject”). Given what >> you’d >> posted earlier I pictured something like this: >> >> GLM y by x1 x2 x3 x4 with cov. >> >> Four between factors with a covariate, all interactions among factors, no >> factor by covariate interactions. >> >> >> >> Gene Maguin >> >> >> >> *From:* SPSSX(r) Discussion < > SPSSX-L@.UGA > > *On Behalf Of *Jillian >> Ma >> *Sent:* Friday, July 27, 2018 3:01 PM >> *To:* > SPSSX-L@.UGA >> *Subject:* Re: Interpreting interactions with covariate >> >> >> >> That makes sense to me. The model does include a significant interaction >> between the two within-subject IVs and the covariate. I might be able to >> explore this with some plotting. I do wonder why by default SPSS does not >> tell me about any interactions between the between-subject IVs and the >> covariate. >> >> >> >> On Fri, Jul 27, 2018 at 2:46 PM, Bruce Weaver < > bruce.weaver@ > > >> wrote: >> >> Does your model include any interactions involving the covariate? If >> not, >> then the nature of the interactions involving your 4 factor variables >> does >> not depend on the value of the covariate, and should be able to proceed >> as >> shown in the Howell & Lacroix article, /I think/. >> >> >> >> Jillian Ma wrote >> > Thanks, Bruce. This resource seems very helpful to try to tackle the >> > LMATRIX and MMATRIX subcommands. I have still not figured out what to >> do >> > now with a covariate. I believe I cannot simply look at the interaction >> at >> > all levels of the covariate because my covariate is a continuous >> variable. >> > Please let me know if you have any clues on what to do with this >> > covariate. >> > >> > On Fri, Jul 27, 2018 at 9:23 AM, Bruce Weaver < >> >> > bruce.weaver@ >> >> > > >> >> > wrote: >> > >> >> Hello Jillian. One option would be to use LMATRIX and MMATRIX >> >> sub-commands, >> >> as shown by Howell & Lacroix (2012). >> >> >> >> http://www.tqmp.org/RegularArticles/vol08-1/p001/p001.pdf >> >> >> >> But personally, I find the requirement of generating all those >> >> coefficients >> >> off-putting, and a likely source of error. Interaction contrasts like >> >> this >> >> are much easier to do in Stata, IMO. E.g., the code might look >> something >> >> like this: >> >> >> >> // ANCOVA model with x as continuous covariate >> >> anova y c.x a##b##c##d >> >> // a*b*c interaction at each level of d >> >> contrast a#b#c@d, effects nowald >> >> // a*b interaction for each c*d combination >> >> contrast a#b@c#d, effects nowald >> >> // Effect of a for each b*c*d combination >> >> contrast a@b#c#d, effects nowald >> >> // Etc. >> >> >> >> >> >> Jon, if you're looking for extension command ideas, something that >> >> performs >> >> interaction contrasts like this more easily would be a good one, IMO. >> >> >> >> HTH. >> >> >> >> >> >> Jillian Ma wrote >> >> > Hi, >> >> > >> >> > I have several dependent variables for which I have conducted >> 2x2x2x2 >> >> > ANCOVAs with the same IVs and covariate using the GLM command. In >> >> several >> >> > of these ANCOVAs, there are significant 4-way interactions that I am >> >> > trying to interpret. It is my understanding that I should try to >> tease >> >> > apart the effects by splitting the file by one of my IVs and working >> my >> >> > way down the lower order interactions to see where the effects are. >> >> > >> >> > However, I am lost with regard to the covariate. Should I be trying >> to >> >> see >> >> > the effects solely at the the mean of the covariate? How can I >> describe >> >> > the patterns when they are likely changing at different values of >> the >> >> > covariate? >> >> > >> >> > Additionally, is there a reason why SPSS does not test for >> interactions >> >> > between the covariate and between-subject IVs? And how do I >> interpret >> >> an >> >> > interaction between my within-subject IVs and my covariate? I have >> >> thought >> >> > about plotting my covariate along the horizontal axis with different >> >> lines >> >> > for the levels of an IV, but this is not as simple as seeing that >> there >> >> > are different slopes. >> >> > >> >> > Any advice would be so greatly appreciated! >> >> > >> >> > Jillian >> >> > >> >> > ===================== >> >> > 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 >> >> >> >> >> >> >> >> >> >> >> >> ----- >> >> -- >> >> Bruce Weaver >> >> >> >> > bweaver@ >> >> >> http://sites.google.com/a/lakeheadu.ca/bweaver/ >> >> >> >> "When all else fails, RTFM." >> >> >> >> NOTE: My Hotmail account is not monitored regularly. >> >> To send me an e-mail, please use the address shown above. >> >> >> >> -- >> >> Sent from: http://spssx-discussion.1045642.n5.nabble.com/ >> >> >> >> ===================== >> >> 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 >> >> >> > >> > ===================== >> > 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 >> >> >> >> >> >> ----- >> -- >> Bruce Weaver >> > bweaver@ >> http://sites.google.com/a/lakeheadu.ca/bweaver/ >> >> "When all else fails, RTFM." >> >> NOTE: My Hotmail account is not monitored regularly. >> To send me an e-mail, please use the address shown above. >> >> -- >> Sent from: http://spssx-discussion.1045642.n5.nabble.com/ >> >> ===================== >> 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 >> >> >> >> ===================== 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 >> > > ===================== > 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 ----- -- Bruce Weaver [hidden email] http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." NOTE: My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. -- Sent from: http://spssx-discussion.1045642.n5.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
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
Re: Interpreting interactions with covariate
|
|
Ah, I see now, thank you. Below is the full syntax I used: GLM mrimp mriip mrbmp mrbip BY readspint sex WITH vscore /WSFACTOR=interest 2 Polynomial difficulty 2 Polynomial /METHOD=SSTYPE(3) /EMMEANS=TABLES(OVERALL) /EMMEANS=TABLES(interest) /EMMEANS=TABLES(difficulty) /EMMEANS=TABLES(sex) /EMMEANS=TABLES(readspint) /EMMEANS=TABLES(interest*difficulty) /EMMEANS=TABLES(sex*interest) /EMMEANS=TABLES(sex*difficulty) /EMMEANS=TABLES(readspint*interest) /EMMEANS=TABLES(readspint*difficulty) /EMMEANS=TABLES(readspint*sex) /EMMEANS=TABLES(sex*interest*difficulty) /EMMEANS=TABLES(readspint*sex*interest) /EMMEANS=TABLES(readspint*sex*difficulty) /EMMEANS=TABLES(readspint*interest*difficulty) /EMMEANS=TABLES(readspint*sex*interest*difficulty) /PRINT=DESCRIPTIVE ETASQ PARAMETER HOMOGENEITY /CRITERIA=ALPHA(.05) /WSDESIGN=interest difficulty interest*difficulty /DESIGN=readspint sex readspint*sex vscore. On Fri, Jul 27, 2018 at 5:08 PM, Bruce Weaver <[hidden email]> wrote: To give the info Gene is asking for, you need to include the complete syntax |
Re: Interpreting interactions with covariate
|
Administrator
|
Thanks for posting that. One of the drawbacks of using GLM is that it ties
your hands pretty severely as to which terms are included in the model. That is one reason why it is often preferable to restructure the data from WIDE to LONG and to then use MIXED to estimate the model. Not only does it allow you more flexibility re the covariance structure, but it also allows you complete control over which terms are included in the model. GLM pretty much ties your hands as to which terms are in the model, as the following example shows. HTH. * Get some data for 2x2x2x2 model with repeated measures on the last two factors. * Modify path to folder with SPSS sample datasets. GET FILE='C:\SPSSdata\survey_sample.sav'. * Let A and B be between-Ss factors. COMPUTE A = wrkstat EQ 1. COMPUTE B = marital EQ 1. FORMATS A B (F1). DESCRIPTIVES A B. * Use the 4 education variables (educ paeduc maeduc speduc) for the repeated measures. * Use age as the covariate. COMPUTE X = age. * Use default settings from the GUI. GLM educ paeduc maeduc speduc BY A B WITH X /WSFACTOR=C 2 Polynomial D 2 Polynomial /WSDESIGN=C D C*D /DESIGN=X A B A*B. * Between-Ss effects: X A B A*B Error. * Within-Ss effects: C C * X <-- No way to exclude this term C * A C * B C * A * B Error(C) D D * X <-- No way to exclude this term D * A D * B D * A * B Error(D) C * D C * D * X <-- No way to exclude this term C * D * A C * D * B C * D * A * B Error(C*D). * NOTE that all interactions involving the categorical variables are included; * and all interactions of covariate with repeated-measures factors are included. * There is no way to tell GLM to exclude those interactions. * We can exclude the C*D and A*B interactions though. * Remove the two-way interactions from the WSDESIGN & DESIGN sub-commands. GLM educ paeduc maeduc speduc BY A B WITH X /WSFACTOR=C 2 Polynomial D 2 Polynomial /WSDESIGN=C D /DESIGN=X A B. * Between-Ss effects: X A B Error. * Within-Ss effects: C C * X <-- No way to exclude this term C * A C * B Error(C) D D * X <-- No way to exclude this term D * A D * B Error(D). * As shown ablve GLM allows you very little control over which terms * are included in the model, and no control over interactions of covariates * with within-Ss factors. * If you restructure the data from WIDE to LONG and use MIXED * to estimate your model, you will be completely free to determine * which terms to include in the model. Jillian Ma wrote > Ah, I see now, thank you. > > Below is the full syntax I used: > > GLM mrimp mriip mrbmp mrbip BY readspint sex WITH vscore > /WSFACTOR=interest 2 Polynomial difficulty 2 Polynomial > /METHOD=SSTYPE(3) > /EMMEANS=TABLES(OVERALL) > /EMMEANS=TABLES(interest) > /EMMEANS=TABLES(difficulty) > /EMMEANS=TABLES(sex) > /EMMEANS=TABLES(readspint) > /EMMEANS=TABLES(interest*difficulty) > /EMMEANS=TABLES(sex*interest) > /EMMEANS=TABLES(sex*difficulty) > /EMMEANS=TABLES(readspint*interest) > /EMMEANS=TABLES(readspint*difficulty) > /EMMEANS=TABLES(readspint*sex) > /EMMEANS=TABLES(sex*interest*difficulty) > /EMMEANS=TABLES(readspint*sex*interest) > /EMMEANS=TABLES(readspint*sex*difficulty) > /EMMEANS=TABLES(readspint*interest*difficulty) > /EMMEANS=TABLES(readspint*sex*interest*difficulty) > /PRINT=DESCRIPTIVE ETASQ PARAMETER HOMOGENEITY > /CRITERIA=ALPHA(.05) > /WSDESIGN=interest difficulty interest*difficulty > /DESIGN=readspint sex readspint*sex vscore. ----- -- Bruce Weaver [hidden email] http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." NOTE: My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. -- Sent from: http://spssx-discussion.1045642.n5.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
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |