Hi everybody
In our study, we have 3 groups of subjects: personality disorder patients (G1), depression/anxiety patients (G2), and non-clinical subjects.
All of them filled a questionnaire characterizing themselves (SP: self-description). A close family member filled the same questionnaire characterizing the subjects (HP: hetero-description), and the subjects filled again the same questionnaire guessing their characterization from the family member (GP: guessed-description).
Scores are in scale format.
G1 had more and higher Pearson correlations between SP-HP and HP-GP than the other groups. Non-clinical subjects had fewer and weaker correlations than the other two groups.
The first publication on this is here:
https://www.researchgate.net/publication/267156291_P-17-006-Does_personality_disordered_patients_view_themselves_as_others_view_them
Just counting and describing correlations doesn't seem to me the better way to analyze the data. What do you think is the best suitable model of analysis for comparing the groups?
Thank you all,
Helena
Conference Paper P-17-006-Does personality disordered patients view themselve...
See matarials of gonference
http://eventegg.com/esspd-borderline-2014/
Maybe multiple regression could work for this? For example, if you let HP be your dependent variable and let your independent variables be SP, Group, and the SP*Group interaction, then you can both test whether there is an association between HP and SP scores overall (the SP main effect) and whether that association differs between groups (the interaction). You would probably want to look at standardized rather than unstandardized coefficients.
This might not be perfect because you still then would have to run a second regression model to look at the association between HP and GP (and possibly a third if you also care about the association between SP and GP). It might be possible to do the whole t hing in a multivariate regression, but I'm not too familiar with those so I'm hesitant to say anything about it; you could look into it, though.
Others on the list might know of something more appropriate than regression for this.
Hi Helena,
what about using Fisher's z test to compare correlation coefficients?
e.g., http://vassarstats.net/rdiff.html
Although I see that it would imply repeating it for each pair of correlation, so I am pretty sure a multivariate technique would be more appropriate (but is there enough power?)
Best,
Carlo
Helena depende do tipo de output. a resposta acerca da patologia é dada como? dicotomica? likert?
Hi Stephen, Thank you for your answer. You gave me something to think about.
Best, H
Hi Carlo and Renzo, my problem is the risk of type I error. But power is an issue indeed. Many thanks
Olá Margarida, obrigada pela pergunta.
Através do questionário que administrámos aos 3 grupos nas "3 situações" [auto (SP), hetero (HP) e adivinhada (GP)] temos pontuações para várias dimensões, sendo cada dimensão uma variável em formato escalar.
Como mostrar que o maior número e que a maior força destas correlações em G1 é significativamente diferente dos outros grupos?
Olá Helena, se bem entendi julgo que o melhor será o modelo de equações estruturais precedido de uma AFE. Vê a tese de doutoramento em https://repositorio.iscte-iul.pt/bitstream/10071/6590/1/TESE_FINAL_ANA_MARTINS_Junho_2013.pdf
se queres uma coisa mais simples recorre a uma análise de medidas repetidas
bom trabalho
- Badges
- Science topic