regression

Thus: reference group is formulated by taking everything into account which is not listed in the term collumn

Note: access to the full survey

User

improved a lot

is much better if we use ordinal model even if high chi2

no scalar invariance even if we keep the constraint

another change

impose constrain to improve the model

meaning This item work differently between male and females

the highest, look up what p64 is

even if we freeze some of the thresholds, we will not reach an improvement of chi2 in our model no scalar invariance possible

we are unable to compare the thresholds

a model with covariance included fits the data much better

exclude the covariance

the model is similar, but you can see that some SE are actually different!

we losed 2

tells you how the variable is distributed

for each of teh variables that you have

not estimated since it does not make sense

loadings are better now compared to the continuous case

Item response theory to build a model which uses only binary data

the estimates of the factor loadings will be the same when taking non robust version the point estimate is the same while standard error and residual variances do change

again we see robust version

robus version is lower, which is good

the other two are the scale versions

we improve the fit indices here

the test statistic is lower, which is good, we are improving the model, we scale the variance-covariance matrix with a kurtosis, see how big it is underneath

reflect the scaling factor with which the indices are corrected

fit statistic and standard error become more robsut

These estimators makes it possible to still run the model it can be seen as a scaling procedure by its kurtosis it panalize certain variables more than other

here there is non normality

this is the joint distribution of all the variables included in the dataset

Here we see that our variables are non of them are normally distributed

You should run such test for all the varaibles which are continuous

You can actually test for any variable if it follows a certain distribution by changing this parameter

Is a normal test to see if there is non-normality

What is this? People tend to agree that well state response only old etc don't agree on this

It is easy to just plot the variable to see if it is normally distributed or not, underneath we see that this is not the case

Usually with survey we prefer skewed distribution

it is boarder line significant

not significantly different, since the coefficients are very similar thus the two models are not that the different even though the path is different, it is not informative

the model can be improved

this is different, yet the coefficient is different, we do not really want to take this into account

this need to be written by hand (the equation

conduct an additional test, to see if it is the case

indicate the difference will the model improve if you let the intercept to be estimated freely

this is the biggest

we do not reach the scalar invariance

mean estimated the females mean is higher a bit compared to the males, yet this is not that big for this you can then use for instance a t-test to see if there is a significant difference between the groups

mean 0

to identify the model, you need to set the mean of one group to 0

here we can compare the means across the groups

estimated to be the same across the groups

not able to do anything, not identified

not necessary to allow the path to be different since they are similar

even though the loadings are the same for the two groups but the regression can be different (other paths)

specify the path model recall it is the same as previous session

In OLS we just add an interaction you want direct and indirect effects are different across groups

here we have structural path is our path coefficient is different across different groups?

test

overview of all parameters tells you the meaning of the above the parameters in the table

this is the contribution of letting it free over all chi2

this is the boarder line

it tells us whether allowing something to be free across the two groups (residual e.g.), will improve the chi2 of our model?

at least 2 indicators need to be the same across the groups afterwards we can only maybe change it

the model fits very well even though there are difference

we want to test if the data does not get tooo bad when we can't use it anymore

even though fit indices are good this means that the model, it reject the null hypothesis the closer to 0, the better see big decreases (chi2) everytime we modify our model, what does it mean technically? Does the model improve if we put constraint? it becomes more simple, we want to see good model ==> not bad

Very good indices, no! There should be another test, since this test is not powerful enough, look at likelihood ratio test

tells lavaan fix the coefficient to be the same across the two groups

Variance is the same (assumption)

vector of 2 if we reach this type of invariance, a group is not systimatically expected to give the same response

the meaning might be that the interpreation of the question is different for the two groups or same response style

Here we want the loadings to be the same

this is lower compared to males are they statistical significant different or are they the same?

Transforming into a factor, the two covariance matrices are independent

useful when you have strong theoretical background and scales, then you want the variance or residuals to be the same in certain cases

across the different groups

• when you want to compare the mean
• research question: how do groups differ compared to the other group? (same as t-test)

in each group

• processing of testing, if latent construct is understand the same way across the groups can be applied to different grouping shema Intense if a lot of group, since you need to understand each loading for each group (can be a lot of work)

amount of variance explained in the model

these are the regression paths this is a way to extract the code and put it into a table

this is better compared to the previous model, since we identified more mediation paths, note if you add mediation paths this needs to be justified by theory

only 1 path, since we are estimathing one prediction

how good is your prediction model? you can also use R^2

only look for measurement model not for regression path

as more covariates are introduces the model get worse, estpecially cfi and tli decrease a lot why for cfi and tli? you get bigger variance and covariance matrix, thus generating more error, since more variables are getting correlated (see extra paper note)

good, even though negative factor loading, your fit is good, thus fit model does not tell you anyting about the loadings

this is wrong, delete

the text underneath is related to this section

indirect effect is significant and negative which pass through mediation taking it alone is not significant

1 point increase in your elgolatorism (1sd) you have 0.20 decrease on wellfort support

here this is interesting from positive to negegative

income has no effect on well fort

terrible latent factor, either you need to reverse or this indicator does not work thus kick it out, thus before you should have seen it and deleted it

again calculating a new paramer

stands for new parameter ab is just a name it can be called anything else it tells you the coefficients, multiply it together and give the result in the label specified

this is a label to indicate how the regression path is called we do exactly the same for path a and b

how much the mediation explain the effect from x to y

You are not required to use latent factors to do such analysis, you can also use manifest variables you can even fit multilevel mediation models It is important to have theory behind such analysis

yet these are not significant, so not reporting in the paer

1sd in education of our respondent correspond on average 0.010 factor

look at this when you are working with dummy variables not standardized

dummy