Hello,
I have a set of independent qualitative variables related to households (type of house, type of windows, presence of pets, ...), and dependent quantitative variables (number of arthropods collected in the households, species richness and diversity indexes).
Which statistical test would you recommend to understand which factors are affecting the most the arthropods population?
So far I came across Factor analysis of mixed data (FAMD) but these analyses are new for me so I would like to have opinions from people with more expertise than me, thanks a lot for your help.
You could use factor analysis with your interval-level dependent variables to create a single scale, if you think they are all measures the same underlying concept and are thus highly correlated measure of essentially the same thing.
Then you could test regression to test the effects of your independent variables.
Article FACTOR ANALYSIS OF MEDICINE RETAIL IN MACEDONIA
You can look at the work of this link, in terms of creating independent variables and applying factor analysis.
I hope this link is very useful for you
https://www.researchoptimus.com/article/factor-analysis.php
I think the first two words in your question may have lead people to think this question was primarily about how to do a factor analysis with categorical variables. I'm not so sure.
Before suggesting something let me make sure I understand. You have a set of qualitative/categorical variables and you believe that these can be accounted for by a smaller number of dimensions or factors. And you want to use these factors to predict a set of quantitative variables (and presumably not wanting to reduce the dimensionality of these ... i.e., not use something like factor analysis or principle component analyses for these). Is that right? If you just want to compare one set of variables with another set, then there are approaches (e.g., within the GIFI procedures which are part of many stats packages like R, SPSS, ... Article OVERALS: Nonlinear canonical correlation with k sets of variables
).
Hi Daniel Wright and thanks for your help.
Apologies, my question was not very clear and I hope I am not oversimplifying a more complex matter. I have a set of independent categorical variables and dependent quantitative variables. I would like to know which test is more suitable for understanding which independent categorical variables are associated with changes in the dependent quantitative variables. Thank you.
Thank you for clarifying the question, Federica Boiocchi. Regarding the dependent variables, do you have multiple univariate questions or one multivariate question? Putting it another way, do you wish to examine each DV individually, or are you content with looking at one "optimal" linear combination of all of the DVs? See the classic article by Carl Huberty and John Morris for more info. HTH.
Article Multivariate Analysis Versus Multiple Univariate Analyses
Thank you Bruce Weaver ,
My original plan was to look at all the variables together. Although, I did not think about this aspect and my research is exploratory so I'm wondering if examining each dependent variable comparing them individually to the dataset of independent variables may be useful, I imagine in this way I could obtain a value of the association between DV and IVs? Do you have any suggestion? Thank you very much.
If you can combine all the measures of the dependent variable into a single measure, that would give you the clearest picture of the outcome. Further, combining the separate indicators into an overall scale helps improve the reliability of that scale (i.e., its ability to detect an effect).
Because there have been so many questions here on this topic, I have compiled a set of relevant resources:
https://www.researchgate.net/post/How_can_one_help_people_create_simple_scales_from_Likert-scored_items
Hi Federica Boiocchi , looking how one set of variables is associated with another set is done for a fairly different purpose than saying how associated each individual variable in set A is associated with each individual variable in set B. The former would you would use a flexible form of canonical correlation (e.g., http://deleeuwpdx.net/janspubs/1994/articles/vanderburg_deleeuw_dijksterhuis_A_94.pdf ) and come up with a set of functions that show how responses in set A can be combined and responses in set B can be combined to create the highest association. The latter would, the bivariate analysis, is more straight forward. You might produce a table of these bivariate associations (could be eta^2, omega^2 from ANOVA, or something else). Because you'd have lots of these you'd want to take into account multiple tests. A table with these effect sizes and their intervals would be a good first step.
As far as which of these is more appropriate for your research, that depends on your research. I can say that I have often seen people jump into the multivariate procedures because they have multiple variables, even though their research questions dictate that they should have been looking at bivariate procedures. If I were a betting person (I live in Vegas!), I'd bet the bivariate method is more appropriate for your exploratory research.
Thank you very much David L Morgan and Daniel Wright for your help!! It is very appreciated :)
- Badges
- Science topic
- Similar topics
- Biological Science
- Zoology
- Invertebrate Zoology
- Arthropods