Dear Arvind Negi,
You have various options in analysing your data. A suitable multivariate technique can be the MCA\ Multiple Correspondence Analysis (if your data are qualitative in nature) . In this sense this technique can be used to explore the data matrix of the variables considered (the four independent variables plus more variables). In this way you can explore the structure of your variables and eventual relationships you can encounter. Please consider is an exploratory technique.
Then you can apply a Clustering Method (K-Means, or a Hierarchical Clustering Method) in order to group the statistical units (the rows) in an useful way. In this sense you can obtain interesting ideas in your data exploration both considering the columns of the data matrix (which show the relatinships between the variables) and the rows (which show interesting groups of observations or cluster).
In my opinion you can to maximize the information in your data analysis by increasing the number of your columns and rows. In this sense you can use more than 4 qualitative variables.
If you need more references you can consider this article:
Abdi, H., & Valentin, D. (2007). Multiple correspondence analysis. Encyclopedia of measurement and statistics, 651-657.
A free implementation of the statistical method is in the package FactoMineR. Here you can find more informations on the way to concretely apply the method.
http://factominer.free.fr/
Kind Regards,
Carlo Drago
Dear Arvind, Which kind of relationships? If not specific consider them 4 random variables then you have many analysis techniques to relate them.
Dear Arvind Negi,
I think I'd need more information regarding your variables to have more success in helping you. However, if I assume that the independent variables you want to test are continous, I would perform a correlation to assess the dependence between them.
The most commonly used measure is the Pearson's correlation coefficient, which is estimated by dividing the covariance of n variables by the product of their standard deviations.
If you think, by any reason, that you'll have to work with non-parametric statistics, you have another ranking measures, like Spearman's rank correlation coefficent and Kendall tau rank correlation coefficient. These coefficients will show you extent to which, as one variable increases, the other variable tends to increase, without requiring that increase to be represented by a linear relationship.
Hope that this brief explation is useful for you.
Best regards,
Pedro.
Peter talking about the type of variable and no real relationship between the variables. If these are independent of each other (really), no significant correlation ought to be, for more method that you occupy.
Moreover, if you mention the method to be applied to observe the existence of a possible relationship between these variables. As you write, dependent variable.
Pearson, Spearman, Kendall Tau, V Cramer dependence Ji Square, or another.
If you have a linear model or not, could be the way multicollinearity.
best regards CJ
I have four different categories containing different data set based on clinical studies (1. Age of the people (Independent), 2. Symptoms (independent), Different regions of india(Independent), Total people affected in that region the same disease (again independent ). So I want to applying a correlation or dispersion stats on these results.
Yes there is use the suitable multivariate technique like clustering to find out some relation between them or find out correlation or other alike measures
@ Kaustav thank you for the answer. But can you provide me some more information so that i have better understanding.
suppose you have information of 4 variables corresponding to 4 independent traits of any organism and you want to find some relationship between these traits from that data. For this the best way to find some relation is to go for hierarchical clustering method and you will have a dendogram which will help you to find the similarity between those variables if there exist some kind of relationship.
Try PCA. It find patterns of covariance across multivariate data. It's available in virtually all math/stats packages.
If you have some variables in which dependent and otherwise independent variables then we study the correlation by cannonical correlation
Dear Arvind Negi,
You have various options in analysing your data. A suitable multivariate technique can be the MCA\ Multiple Correspondence Analysis (if your data are qualitative in nature) . In this sense this technique can be used to explore the data matrix of the variables considered (the four independent variables plus more variables). In this way you can explore the structure of your variables and eventual relationships you can encounter. Please consider is an exploratory technique.
Then you can apply a Clustering Method (K-Means, or a Hierarchical Clustering Method) in order to group the statistical units (the rows) in an useful way. In this sense you can obtain interesting ideas in your data exploration both considering the columns of the data matrix (which show the relatinships between the variables) and the rows (which show interesting groups of observations or cluster).
In my opinion you can to maximize the information in your data analysis by increasing the number of your columns and rows. In this sense you can use more than 4 qualitative variables.
If you need more references you can consider this article:
Abdi, H., & Valentin, D. (2007). Multiple correspondence analysis. Encyclopedia of measurement and statistics, 651-657.
A free implementation of the statistical method is in the package FactoMineR. Here you can find more informations on the way to concretely apply the method.
http://factominer.free.fr/
Kind Regards,
Carlo Drago
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics