Date: Fri, 18 May 2012 11:34:23 -0700
Reply-To: David Marso <david.marso@gmail.com>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: David Marso <david.marso@gmail.com>
Subject: Re: Subject specific information
In-Reply-To: <7DE1AA9F-BFAA-4FA0-88E0-EA50B0ED2836@PHHP.UFL.EDU>
Content-Type: text/plain; charset=UTF-8
How do you get 32 rows from 3 dichotomies?
I suspect there would be 8 (or you have two other dichotomous variables you
are not talking about).
I do hope you have counterbalanced or randomized your target1..targetk
combinations over time or you have a big mess.
I will not provide specific syntax but I would direct you to study the MIXED
procedure.
However I believe it is designed to fit univariate responses but it permits
flexibility in specifying error structures unlike GLM.
----------------------------------
Craggs,Jason G wrote
>
> Sorry for the confusion. I was trying to convey that the same subject
> rates multiple dimensions of all possible combinations of the dichotomous
> variables. For example:
> Time ID Target1 Target2 Target3 rate1
> 1 1 0 0 0. 30
> 2 1. 0. 0. 1. 62
> 3. 1. 0. 1. 0. 15
>
> Etc. etc.
>
> I will look into the split file option. Thank for the suggestion.
>
> Jason
>
> On May 17, 2012, at 10:57 PM, "Rich Ulrich"
> <rich-ulrich@<mailto:rich-ulrich@>> wrote:
>
> 1. Oh. That is contrary to your example, which shows all
> the dichotomous values as the same for a subject, row 1, 2, [ ,], 32.
>
> I think you can get the information you want on parameter
> estimates from repeated measures, but I never have.
>
> 2. Split Files certainly generates the sets more readily than
> using a bunch of Select If's. The new problem might be in
> preserving the IDs for the sets.
>
> --
> Rich Ulrich
>
> ________________________________
> From: jcraggs@.UFL<mailto:jcraggs@.UFL>
> To: rich-ulrich@<mailto:rich-ulrich@>
> CC: spssx-l@.uga<mailto:spssx-l@.uga>
> Subject: Re: Subject specific information
> Date: Fri, 18 May 2012 02:41:34 +0000
>
> Hi Rich,
>
> Thanks for the reply.
>
> 1. All data are nested within subject. The three dichotomous variables
> refer to characteristics of a target in a vignette (e.g., young/female/AA)
> that may influence the subsequent ratings. Basically we're looking for
> idiographic decision policy/rating biases.
>
> 2. The current analysis is scripted to basically "select if" for each
> subject and perform a regression. However, there are now several subject
> groups and will eventually compare these as well. Using an MLM approach,
> with ID being a random effect, I can get some of the information I want,
> but am stuck as how to get the rest. Getting subject specific betas and
> looking at interaction effects being the two biggest conundrums at the
> moment.
>
> Cheers,
> Jason
>
> On May 17, 2012, at 10:18 PM, "Rich Ulrich"
> <rich-ulrich@<mailto:rich-ulrich@>> wrote:
>
> 1. All three of your dichotomous IVs are between-subject
> variables; the within-subject trials are irrelevant to their
> main effects and interactions. So you will simplify this
> analysis if you aggregate across rows, and look at the
> easy 2x2x2 ANOVA.
>
> It is more complicated if you also want to look at effects
> and interactions with "32 observations" in some fashion.
>
> 2. For subject-specific effects across the 32 observations,
> it appears that you are looking for the linear trend/ linear
> regression. I would probably use Split Files and Regression,
> but if you are looking for immediate tests on the within-effects,
> there probably are direct ways to get them. Is Regression
> enough for you?
>
> --
> Rich Ulrich
>
>
> ________________________________
> Date: Thu, 17 May 2012 20:36:51 +0000
> From: jcraggs@.UFL<mailto:jcraggs@.UFL>
> Subject: Subject specific information
> To: SPSSX-L@.UGA<mailto:SPSSX-L@.UGA>
>
>
> Greetings,
>
>
>
> I’ve a dataset organized for HLM/MLM analyses (each subject spans several
> row, 1 row per observation, for 32 observations).
>
> Specifically, the data are organized as such:
>
> Time ID TargetAge TargetSex
> TargetRace Rate1 Rate2 …
>
> 1 1 [0,1] [0,1]
> [0,1] [0-100] [0-100] …
>
> 2 1 [0,1] [0,1]
> [0,1] [0-100] [0-100] …
>
> … 1 [0,1] [0,1]
> [0,1] [0-100] [0-100] …
>
> 32 1 [0,1] [0,1]
> [0,1] [0-100] [0-100] …
>
> 1 2 [0,1] [0,1]
> [0,1] [0-100] [0-100] …
>
> 2 2 [0,1] [0,1]
> [0,1] [0-100] [0-100] …
>
> … 2 [0,1] [0,1]
> [0,1] [0-100] [0-100] …
>
> 32 2 [0,1] [0,1]
> [0,1] [0-100] [0-100] …
>
> 1 … [0,1] [0,1]
> [0,1] [0-100] [0-100] …
>
> … … [0,1] [0,1]
> [0,1] [0-100] [0-100] …
>
> 1 300 [0,1] [0,1]
> [0,1] [0-100] [0-100] …
>
> … 300 [0,1] [0,1]
> [0,1] [0-100] [0-100] …
>
> 32 300 [0,1] [0,1]
> [0,1] [0-100] [0-100] …
>
>
>
> TargetAge coded: young/old
>
> TargetSex coded: male/female
>
> TargetSex coded: AA/Cauc
>
>
>
> The goal is to use the first three “target” variables as the independent
> variables and the subsequent ‘Rate…’ variables as the dependent variables.
>
> While I am familiar with running fixed and random effects analyses, and
> saving and plotting the estimated predicted values, I am struggling to
> accomplish two goals using SPSS-20.
>
> 1. I would like to test the main- and interaction- effects of the
> IV’s. Testing the main effects seems fairly straightforward, but I am
> unclear as to the best way to estimate and test the interaction terms.
>
> 2. I would like to save subject specific values for: intercepts,
> slope, and a standardized beta weight from each of the tests mentioned
> above.
>
>
>
>
>
> Any thoughts, suggestions, and or comments are greatly appreciated.
>
>
>
> Best regards,
>
> Jason
>
--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Subject-specific-information-tp5711695p5712067.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.
=====================
To manage your subscription to SPSSX-L, send a message to
LISTSERV@LISTSERV.UGA.EDU (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
| 
