Hello,
I've got a quite complicated hierarchical analysis to conduct. I would like to ask for an advice how to input my data to mixed models. I have 100 respondents answering a question about trust towards persons, whose descriptions are a combination of an adjective and a noun. A combination may be coherent or not (my main independent variable). f.ex: How much would you trust... a tall basketball player (coherent) a short basketball player (incoherent) a tall dwarf (incoherent) a short dwarf (coherent) ... and I have many of such 4-element clusters of questions. another example, to better grasp the problem: fat model (incoherent) skinny model (coherent) fat cook (coherent) skinny cook (incoherent) Now, apart from the random effect of a respondent, with which I know how to deal with, I have random effects of both the nouns and the adjectives. That is, people may for some reason trust more a model than basketball player, or a tall person than a short person, so I want to get rid of these effects and study only the effect of COHERENCE of each combination. So I thought of two possibilities, how to deal with this random variance: 1) one can substract the mean for the whole 4-element cluster, and be left with deviations (residuals) of each combination within the cluster 2) (which I think is more legitimate), substract the mean for each noun (2 items) and each adjective (2 items) and have the residuals for each for the coherent/incoherent combination. The problem with this analysis is that what you get at the end are two separate sets of residuals, one for nouns, and one for adjectives. Now, how should I combine them to finally measure trust evoked by each combination?? I thought of a simple sum, but I am not sure.... Any suggestions welcome! Best, Karolina |
Would you please provide more detail so that I (we) can visualize the procedure generating the data. How many of these 4-element clusters did each person get? Did each person get the same set of clusters or did people get 'x' many randomly selected from a huge pool of such clusters? If each person saw the same set of clusters, were the clusters presented in the same order to each person? Did you create these quadruplets by selecting a set of nouns and a set of adjectives and pairing each noun with each adjective or by selecting pairs of nouns and pairs of adjectives, then pairing them up to make the 4-element cluster and never using either of the nouns or adjectives again, or some other procedure?
It seems to me that ignoring clustering, noun effects, adjective effects, presentation order effects, you have a paired t-test: coherent vs incoherent. Are you expecting/wanting to test for noun and/or adjective effects? Gene Maguin -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Karolina Sent: Thursday, October 20, 2016 5:04 AM To: [hidden email] Subject: a complicated HLM problem Hello, I've got a quite complicated hierarchical analysis to conduct. I would like to ask for an advice how to input my data to mixed models. I have 100 respondents answering a question about trust towards persons, whose descriptions are a combination of an adjective and a noun. A combination may be coherent or not (my main independent variable). f.ex: How much would you trust... a tall basketball player (coherent) a short basketball player (incoherent) a tall dwarf (incoherent) a short dwarf (coherent) ... and I have many of such 4-element clusters of questions. another example, to better grasp the problem: fat model (incoherent) skinny model (coherent) fat cook (coherent) skinny cook (incoherent) Now, apart from the random effect of a respondent, with which I know how to deal with, I have random effects of both the nouns and the adjectives. That is, people may for some reason trust more a model than basketball player, or a tall person than a short person, so I want to get rid of these effects and study only the effect of COHERENCE of each combination. So I thought of two possibilities, how to deal with this random variance: 1) one can substract the mean for the whole 4-element cluster, and be left with deviations (residuals) of each combination within the cluster 2) (which I think is more legitimate), substract the mean for each noun (2 items) and each adjective (2 items) and have the residuals for each for the coherent/incoherent combination. The problem with this analysis is that what you get at the end are two separate sets of residuals, one for nouns, and one for adjectives. Now, how should I combine them to finally measure trust evoked by each combination?? I thought of a simple sum, but I am not sure.... Any suggestions welcome! Best, Karolina -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/a-complicated-HLM-problem-tp5733328.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 |
Ok,
here how it was: <<How many of these 4-element clusters did each person get? Each person got 16 clusters, i.e. 64 items to assess. <<Did each person get the same set of clusters or did people get 'x' many randomly selected from a huge pool of such clusters? they all got the same (full) pool of items. <<If each person saw the same set of clusters, were the clusters presented in the same order to each person? No, the items were rotated without any reference to clusters (each person got a random sequence of 64 questions). No nouns or adjectives were repeated. >>Did you create these quadruplets by selecting a set of nouns and a set of adjectives and pairing each noun with each adjective or by selecting pairs of nouns and pairs of adjectives, then pairing them up to make the 4-element cluster and never using either of the nouns or adjectives again, or some other procedure? I had independent judges who invented coherent and incoherent couplings. I asked them: what noun would fit or would not fit with adjectives such as "tall", "short" "etc. I chose the most frequent associations, then I had another sample of judges who reassessed the selected pool according to coherence (how each combination fits together) of the pairs. I chose only the quadruplets that had a large difference in assessments (high/low coherence ratio). >>It seems to me that ignoring clustering, noun effects, adjective effects, presentation order effects, you have a paired t-test: coherent vs incoherent. Are you expecting/wanting to test for noun and/or adjective effects? How would you arrange this paired t-test? For each person a mean assessment of coherent/incoherent pairs? I could forget about the noun/adjective effects, though I think some of them are more/less positive, thus evoking more/less trust. I could also have another study and have judges assess positivity of the nouns/adjectives... |
Thank you.
In your original message: >> I would like to ask for an advice how to input my data to mixed models. Are you asking (a) how to structure your data so that you can do a mixed models analysis, or (b) how to write the syntax to do the analysis, or (c) both? >> How would you arrange this paired t-test? For each person a mean assessment of coherent/incoherent pairs? Think long format: 100 persons, 32 records per person, each record has a coherent score and an incoherent score. In terms of your example: tall basketball player (coherent), short basketball player (incoherent) short dwarf (coherent), tall dwarf (incoherent) Repeated ratings on two objects. Lots of things are ignored. Gene Maguin -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Karolina Sent: Thursday, October 20, 2016 3:09 PM To: [hidden email] Subject: Re: a complicated HLM problem Ok, here how it was: <<How many of these 4-element clusters did each person get? Each person got 16 clusters, i.e. 64 items to assess. <<Did each person get the same set of clusters or did people get 'x' many randomly selected from a huge pool of such clusters? they all got the same (full) pool of items. <<If each person saw the same set of clusters, were the clusters presented in the same order to each person? No, the items were rotated without any reference to clusters (each person got a random sequence of 64 questions). No nouns or adjectives were repeated. >>Did you create these quadruplets by selecting a set of nouns and a set >>of adjectives and pairing each noun with each adjective or by selecting pairs of nouns and pairs of adjectives, then pairing them up to make the 4-element cluster and never using either of the nouns or adjectives again, or some other procedure? I had independent judges who invented coherent and incoherent couplings. I asked them: what noun would fit or would not fit with adjectives such as "tall", "short" "etc. I chose the most frequent associations, then I had another sample of judges who reassessed the selected pool according to coherence (how each combination fits together) of the pairs. I chose only the quadruplets that had a large difference in assessments (high/low coherence ratio). >>It seems to me that ignoring clustering, noun effects, adjective >>effects, presentation order effects, you have a paired t-test: coherent vs incoherent. Are you expecting/wanting to test for noun and/or adjective effects? How would you arrange this paired t-test? For each person a mean assessment of coherent/incoherent pairs? I could forget about the noun/adjective effects, though I think some of them are more/less positive, thus evoking more/less trust. I could also have another study and have judges assess positivity of the nouns/adjectives... -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/a-complicated-HLM-problem-tp5733328p5733345.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 |
Free forum by Nabble | Edit this page |