analyse or code personality traits scores

[Srinivasa Chowdappa]

Hi
How can i analyse or code personality traits scores, particularly Extraversion & Neuroticism from Short form revised Eynsech Personality inventory. I have to study an association between personality traits and burnout. I wonder is there any cut-off points to both traits, so that i can do coding and take it from there.
Many Thanks
Srinivasa.

[carole fantini]

Hi there is no cut-off scores for personality traits (that make no sens as there is no normal extraversion or neuroticism). it's always better to consider these variables as continuous .

[Marcos Almeida]

Hello Srinivasa,

Welcome to the Stata Forum.

Since you said you "have to study an association" between two scores, I suggest you take a look at structural equation models.

In Stata, you can get information on this by typing - help sem - or just start by thumbling through the respective Stata manual.

Hopefully that helps.

Best,

Marcos

[Srinivasa Chowdappa]

Thanks Carole for clarifying it. Have a great year 2017.

[Carlo Lazzaro]

Srinivasa:
I woud also take a look at -help pca-.

[Srinivasa Chowdappa]

ManyThanks Marcos for your valuable input.
Since i am new to research arena, Stata Forum is extremely useful.
I have coded burnout into categorical dichotomous based on the standard cut-off points mentioned in the previous meta-analysis. So burnout being the dependent categorical outcome variable, and two personality variables as continuous independent variables, is it right to use Logistic regression instead of MLR?

Kind Regards
Srinivasa

[Srinivasa Chowdappa]

Thank you for your suggestion Carlo, i feel some concepts are hard to grasp sometimes. please correct me if i am wrong - PCA will NOT CONSIDER the response variable but only the variance of the independent variable when compared to Logistic regression, which CONSIDERS how each independent variable impact on response variable.
In my study, burnout is response variable and neuroticism and extroversion are independent variables. need to do LR and then look into confounding, interactions, Goodness-of-fit and assumptions.

Kind Regards
Srinivasa

[Carlo Lazzaro]

Srinivasa:
thanks for providing more details (the golden rule is to provide them in your very first post concerning the query you're intereseted in).
Yes, you're right about the substantive difference between -pca- and -logistic-; the only minor amendment I would suggest is that -logistic- investigate the impact of each predictor on the dependent variable when adjusted for the other predictors.
Eventually, I do not follow your last sequence of statistical procedures: your model should embed what the literature in your research field says about the topic you're investigating, first. Hence, assumptions validity should come first,as each (regression) model is as good as the assumptions that underpin it.

[Phil Bromiley]

Generally, if you have a continuous variable, it is not good practice to make it binary. You're creating measurement error. For example, assume your variable went from 1 to 5 and you made it into a dummy where 1-3 is 0 and 4-5 is 1. Now, you're throwing away the information that a 3 is very different from a 1, etc. If the field insists on it, I guess you have to do it, but it is a poor practice.

SEM is preferred when you have a measurement model because it can correctly model the measurement process. If you have a binary dv, can use a traditional model like logit or probit, or a SEM model (with the appropriate options for a binary dv). The logit/probit approach does not allow for the possibility of measurement error in the rhs variables, but an SEM approach can do so.