path analysis Flashcards
What is the local fit in path analysis?
the local fit is measures like the b-weights and r-squared for the variables and sig t tests
What is the global fit in path analysis?
goodness of fit and model comparison values
What do squares represent in path analysis?
measured variables in the data set
What do circles represent in path analysis?
latent variables that are implied by the model but not measured, the other measured variables tap into these
What are arrows in path analysis?
show the relationship between variables, double headed ones indicate that variables are related and single headed indicate that one predicts another
What is structural equation modelling and how is it diff from regression?
have multiple dvs in same model, variables can be both iv and dv, do not need to make assumptions about whether variables predict each other, when the program does the analysis, it will show you whether there is a r/s between variables you did not consider
What does the chi-square goodness of fit test show?
whether your model is sig diff from the perfect fitting model
What is the default model and what is the saturated model?
default model is our model, saturated is the perfect fitting model
what is confirmatory factor analysis?
a data reduction method, form of path analysis, reduce large no. variables to a couple factors or latent variables, useful for scale dev, good for construct validity
What situation is factor analysis designed for?
when a and b are related because they share a common cause. factor analysis is used to explore this, the latent variable behind them
What is the exploratory approach in factor analysis?
indicate variables to include but lets spss tell you what ones are related and you can consider whether you want a shared variance (factor analysis approach) or a principal component or total variance apprroach
What are oblique or orthogonal?
oblique is when factors relate to each other and are correlated and orthogonal is when they are not correlated
Why do you constrain factors?
constraining one factor loading for a variable to 1 means that another variable w/ a factor of 2 is twice as strong, it helps quantify the strength of the r/s, model won’t run without it either
What are chi squares problems as a goodness of fit measure?
sensitive to non-normality, favours complexity, sensitive to size of correlations, sensitive to measurement error which can reduce power, sensitive to large sample size which makes more likely to have a type 1 error of rejecting null hypothesis when it is truly null
What are measures of absolute fit?
rmsea, srmr, compare to perfect fitting model, range from 0 to 1, closer to 0, the better
What is a measure of incremental fit?
cfi or tli,, compares to worst fit, want it closer to 1, tli penalises complexity and is more difficult to meet
What is the srmr?
standardised root mean residual, want .06 or less for a good fit or less than .08 for minimally acceptable
What is the rmsea?
root mean square error, accounts for model complexity and sample size, want it to be less than .08 for good fit or .11 or less for small samples
what is the cfi?
.95 or above for good incremental fit or .90 for minimally acceptable
what is the tli?
penalises complexity, measure of incremental fit, compares against worst fit, , want .95 or above or .90 or above for min acceept
What is CMIN?
output name for the chi-square
What is the covariance and how does it differ from correlation?
covariance is unstandardised, correlation is standardised b-weight vs. beta
What is a just-identified, over-indentified, and under-identified model?
just-identified is when all degrees of freedom are used, over-identified is when there’s df remaining, under-identfied is when there is insufficient df to test everything
How many variables per factor do you generally want?
3
An MI of what will sig improve the model?
3.84 or above, use biggest MIs first
What is a nested model?
when one model has all the paths of another plus one or more additional paths
What is a non-nested model?
diff variables across models and the more complex model does not contain all paths from the other, can compare using BIC