In intelligence research, personality research and other research fields of psychology, factor analytic models are used to structurize the variables' jungle. For example: To simplify the variable structure of a personality questionnaire several questions (items), that are supposed to measure the same construct are grouped to one (latent) factor (e.g. extraversion). But does this group of items really measure extraversion? Maybe some of the items "load" on a different factor. And Maybe other aspects of the model have flaws as well... How well a factor model fits the data can be examined by a confirmatory factor analysis.
Keywords: Confirmatory Factor Analysis - Factor Analysis - Psychology - Statistics - Research Methods - Personality Psychology - Extraversion - Conscientiousness - Openness - Neuroticism - Agreeableness - Five Factor Model - Big Five - Intelligence - Carroll
Hello and welcome to the world of statistics and research methods.
Confirmatory factor analysis or how good is my factor model.
Before we start a little warning even though I will try to keep it as simple as possible
I won't go too deep into any statistics and I won't discuss any mathematical equations.
The topic of confirmatory factor analysis still remains quite advanced so if you are a beginner
and you've never heard of variances or covariances or degrees of freedom maybe you should listen to
this episode at a later point in time. Nevertheless I frequently get emails from people asking me
well I wonder whether I should study psychology but I've heard the statistics are really heavy.
Is it really true? The answer is unfortunately yes it's not that easy but as with almost
everything in life you can learn this stuff. It's not magic. But let's begin with this learning
process. Let's get it started. And I want to start with the most important question.
What is it all about? Why do we need confirmatory factor analysis? What is it good for?
Absolutely something. And I'll give you two very prominent examples. The first one
is from intelligence research and the second one from personality research. Let's begin with
intelligence. Here you see a very prominent model of intelligence. It's the factor model
of intelligence by Carroll. And you see there's not just one intelligence. There are different
so-called factors. In this model eight different factors of intelligence. You see fluid intelligence
which is about your reasoning abilities. You see crystallized intelligence which is about general
knowledge. But you also see a factor which is called broad visual perception. And there's a
factor about processing speed. How fast is your brain processing information? And all these factors
are not directly observables. So they are called latent factors. What can be observed
are the answers to the different items. So for example for crystallized intelligence you might be
asked what's the capital of France? Or what is the year Columbus discovered America? So these are
items which load. This is the expression which is used here. Which load on this not directly
observable factor crystallized intelligence. And to do so these two items should correlate
with each other quite high. And on the other hand they should correlate only to a limited extent
with items of the factor fluid intelligence or processing speed. And in fact there is a small
correlation between all these intelligence factors. So somebody who scores high in
crystallized intelligence probably scores high in other dimensions of intelligence as well.
And this is the reason why in intelligence research many scientists today agree with this
or a variation of this factor model of intelligence in which there is a so-called
g-factor. A general factor of intelligence. Which means that all these latent factors
fluid intelligence, crystallized intelligence, general memory and learning and so on. That all
these latent factors load on this general factor of intelligence. Which again is not an observable
variable. It's a latent factor. Because we observe small to medium correlations between
many intelligence items and on a higher level between the latent factors we made up this new
factor. So it's all in our mind you might say. And it just expresses that people being good in one
area are probably also good in another area. But it's just a model. We made this model up. It's a
theoretical idea. It's a hypothesis and a method to confirm this model. To show well if we let
thousands of people take intelligence tests then our data suggests that this model is really
working. One method to show that this model is really good is the confirmatory factor analysis.
How well does the model fit the data? And we cannot only apply confirmatory factor analysis
on factor models of intelligence. But we can also apply it to other models. For example
on the big five model of personality. It's the most influential model in personality psychology.
And the idea is that the personality of a person is well described by locating it on these five
factors. So you can have a high score in extraversion. A low score in conscientiousness.
Maybe you are very open to new experiences. You get a high score in agreeableness if you
get along quite well with your colleagues and friends. And if you worry a lot about well what
might people think about me then you have a high score in neuroticism. And what you see here is
that these five factors are hypothesized not to correlate with each other. Because there are no
arrows between the factors. The only arrows you see here go from the factor to the item. So y1
to y6 are items which indicate extraversion. For example person x likes to be in the center of
attention. Or he likes to go to parties or to dance on the dance floor. And as the big five
factors should not correlate with each other these items, these observed variables should
not correlate with items measuring openness or conscientiousness. And what are the other arrows?
What are these e1, e2, e3, e4? These are the error variances. So all variants of y1 can be explained
by variance which is caused by the factor extraversion. And it can be explained by the
error variance. This is the variance which is not explained by the factor. This is the item specific
variance. So this is the model. And using confirmatory factor analysis we ask ourselves
well is this model correct? Is there really no correlation between extraversion and openness
or conscientiousness and agreeableness? One might expect a correlation. And in fact in
some studies small correlations were observed. So the big five factors are not completely
independent of each other. But all in all the big five model of personality has proved quite
well in the past. But most of the time it was tested in literate, in well-developed countries
with educated participants. And an interesting research question is whether this model applies
to all people in the world. Does it apply to illiterate people living a completely different
lifestyle than we do? And this research question was recently investigated in the study
how universal is the big five? Testing the five-factor model of personality variation
among forager farmers in the Bolivian Amazon. So in this study a lot of these forager farmers
were presented a translated version of the big five factor questionnaire. And then also with
confirmatory factor analysis the researchers looked whether the data fit the five-factor model.
And what they found was no, it doesn't work that well. So maybe people from this clan
called the Chimane, maybe these forager farmers have a different kind of personality structure.
And it seems like the big five factor model doesn't work for them. So this is a good example for
how to applicate the confirmatory factor analysis. So the basic question of the
confirmatory factor analysis is how good is the model? If I have a model with one factor
maybe it's not the best model. Maybe I should go for a two-factor model.
Which in this case as you can see is the better solution. Because in the one-factor model we see
that the two items is hungry for knowledge and is happy to try out new things. Load only small
on the factor extra version. The error variance is gigantic. So only a small amount of variance
can be explained by the factor extra version. Whereas in a two-factor model in which we
introduce a new factor, in this case we call it openness, the two items is hungry for knowledge
and is happy to try out new things. Load on openness much better than they were loading
on extra version in the one-factor model. And therefore as you can see the error variance is
much smaller. By the way what are the loadings? Well the loadings are equivalent to the slopes
in a regression analysis. And if you square the loadings, for example the loading of item one on
extra version, if you square the loading you get 0.36. And if you add this up to the error variance
you get one. So again all the variance of the observed variables can be explained by the
latent factor and the error variance.