Factorial analysis

Factor analysis is a reduction method statistics which aims to explain the possible correlations between certain variables. To do this, taking into account the effect of other factors, which are not observable.
So what this analysis does is reduce. Thus, we take a large number of variables and, through this technique, we manage to reduce them to a more manageable size. To do this, a series of linear combinations of those observed with others that are not visible are used.
The two models: exploratory and confirmatory
We have two ways to carry out this statistical technique, there are clear differences between the two that should be known.
- Exploratory factor analysis: In this case, the objective is to know the latent constructs (which are not seen) to be able to check if they can be valid. Thus, we are facing information of an exploratory type that serves to create a later model, but we do not know this a priori.
- Confirmatory factor analysis: In this case, we are facing a statistical confirmation process. We start from a theoretical model created with the existing literature on the phenomenon studied. Later we contrast it to know its degree of validity.
How to perform a factor analysis
Let's see, in a simple way, how an exploratory factor analysis can be carried out, which is one of the most used in social sciences. It should be noted that the points mentioned below can be selected in statistical programs such as SPSS when performing the analysis.
- Reliability analysis: Normally, Cronbach's Alpha is used, which allows knowing the internal consistency of the model. Values greater than 0.70 are considered acceptable.
- Descriptive statistics: These provide us with basic information about the analyzed data. The mean, the variance, or the maximum and minimum.
- Correlation matrix analysis: These calculations are performed by SPSS. Here we must pay attention to whether the determinant is close to zero. On the other hand, the calculated correlations must be different from zero.
- KMO sample adequacy measure: Allows us to contrast the correlation coefficients. On the one hand, the observed ones, and on the other, the partial ones. It takes values between 0 and 1 and is considered acceptable if it is greater than 0.5.
- Bartlett's test of sphericity: In this case, it contrasts that the correlation matrix is an identity matrix, in which case the analysis could not be done. The estimated Chi square is calculated and, if it is less than the theoretical one, the factorial analysis can be done.
- Analysis of communality: Again it is an indicator of relevance. To be valid, it must take values greater than 0.5.
- Rotated Component Matrix: It is used to extract the eigenvalues that are greater than a value, normally 1. In this way, the reduced factors that represent the variables are obtained. Sedimentation graphs and the matrix itself are used to choose the number.
- Total variance explained: Finally, this analysis tells us what is the total variance explained by the proposed model. Thus, the higher this value, the better the model is at explaining the total data set.
Examples of factor analysis
Factor analysis has many applications in different fields of science.
Let's see some examples:
- In marketing it is widely used when we want to know the will to buy. For example, we analyze various socioeconomic, emotional, or personal variables. Once we have them, we reduce their number with factor analysis and we can better interpret them.
- In accounting we can know which items most clearly affect the obtaining of business profits. Thus, we will know where we must influence more.
- In education we can know the predisposition of a student to a subject. Carrying out certain surveys on its way of studying it, we can obtain a database in which to apply factor analysis.
Leave a Reply