I think there is a need to fuse the sub categories into respective categories, because the categories are made up from sub categories'

Dependent var has more than 2 categories , hence multinomial in nature.

Multinomial Logistic regression is to tried in analysis.

Indp.Var can be included as factors and mediator var as covariate.

IBM SPSS , STATA have modules to carry out this process.

Following link can be followed

https://stats.idre.ucla.edu/spss/dae/multinomial-logistic-regression/

Also there are videos in youtube on this specific topic.

Hope this helps

Bhogaraju Anand Thanks for your response. If I do only variable level analysis, I will not have an insight about which categories are significant. How can I do the analysis to identify the significant categories and subcategories of each variable? I want to give some insight into the categories and sub-categories in my document.

Hello Asma,

Could you possibly articulate your specific research question(s), and give examples of these categories/subcategories for your variables? I think that doing so would increase the likelihood that you'd get recommendations that would work best for you.

Good luck with your work.

Bhogaraju Anand & Bhogaraju Anand I have added some more information and example to my question, please have a look for giving some valuable suggestions.

Hi,

Since all are categories, we cannot apply parametric tests. We need to deal only with frequencies and percentages or proportions..

If you do not expect any interactions between the variables then each of the variables i.e IVs or DVs can be separately made into N x n tables example 2x2 or higher category tables with a row or column percentages and do a Chi-sq test and make conclusions.

on IBM SPSS navigate to Analyze/descriptives/Crosstabs and calculate. you will also find layers where you can enter 3rd and 4th..and so on variables.

Alternatively, you have to navigate to Log-Linear models and calculate after filling in the relevant info.

If you want regression model then follow as in the previous discussion by me i.g multinomial Logistic Regr.

Table 1

Rows: Performance Financial - 0 1

Column; Governance Category _ Board size: 0 1

2x 2 cross tab and calculate Chi-sq value.

A similar thing can be repeated with diff. Row and column categories

Bhogaraju Anand so if I have two dependent variables, how can I get combined result for the two... As my contruct is "Value"(unobserved variable) I measure it with performance and transparency, accountability, trustworthiness and efficiency variables and then measure these with certain categories. Similarly I have construct 'reporting structure' with 2 independent variables( have categories and sub-categories) and a mediating construct 'Stakeholders relationship" with 2 mediating variables( have categories and sub-categories) and 6 control variables. How will I interpret the results for the construct 'value' the construct 'reporting structure' and the construct 'stakeholders relationship then?

First Step: Dimensional Reduction of IVs

https://www.youtube.com/watch?v=EDPvpLKjCJU

This video is on Multiple Correspondance analysis reduction of Ivs into groups

Second step; Each of the groups after Multple corr ana can be coded.

Step Three: Take one Dv at a time and run Multinomial Logistic regr. Ana with the covariates as Stakeholders , other variables.

Try this and see

As the DVs are not correlated they can be analyzed separately

Refer to this video below

https://www.youtube.com/watch?v=1BL5cL8_Cyc

or

https://www.youtube.com/watch?v=zRDtkGC4Y5o

Try then and you can find a way.

Pls. keep us informed

Can we also reduce dimensions of DV or DVs similarly by multiple corre. ana?

Bhogaraju Anand Thank you very much for your time. I will let you know once I apply these procedures.

Bhogaraju Anand I need to present all individual variables in regression whether significant or not. If I apply MCA, I will reduce variables. This may limit the predictive capability of the regression model due to reduction of variables. Is there any approach to use all variables during regression and do something step-wise to reduce non-significant variables. Even then, it can reduce categories, it can not reduce variables. I still expect to have all variables in place. For the concept 'value', there will be 5 DVS, For the concept 'reporting structure' I will have 2IVS, for the concept 'stakeholders relationship, I will have 2 mediating variables and then the control variables. If I have these 5 DVs for the concept value, then is it possible to interpret the concept 'value' with multinomial logistic regression. As these will be concepts and variables as proxies, does logistic regression allow these all DVS and IVs in one equation for simultaneous analysis to interpret the results for the concepts, not the proxy variables..... as it happens in SEM which allows interpretation of unobserved constructs(concepts)

Dear Asma Malik ,

I strongly agree with David Morse , the best approach to recommend a potentially most suitable analytical procedure and statistical method(s) is when one gets guided by the specific research questions and research questions you formulated for your study.

All the best!

WOW mindboggling and very confusingly interesting scenario. I strongly feel that you need to reconsider your proposed study and build it on the foundation prepared with the help of Theory and Literature. All variables are categories and sub-categories, which according my understanding is little difficult to comprehend even. However, what Prof. Bhogaraju Anand trying to explain makes sense. Thank you.

If you want to include all IVs , then include all subcategories as IVs so that you will have a total of 20 IVs. The rationale is each of these is a binomial var. Later, after the analysis, they can be grouped into categories if you like because categories are made up of subcategories.

Each DV with several nominal levels perform Multinomial logistic regr. with covariates

test whether DVs are correlated or not by Tetrachoreic or Polychoric correlation. If they are separate (not Correlated) then include them separately. If there is a high correlation between them then treat them as one.

Bhogaraju Anand Thanks for your support. I will come back with the outcome after applying these.

Bhogaraju Anand Hi Anand, I need your help to know a little bit more. Which reliability and validity tests should I use for the data I have. All this data is extracted from the reports. I have done content analysis of the reports to extract this data and then converted to nominal( percentages) except the control variables. Control variables are continuous type but also percentages or nominal

Hi,

There are many types of Reliability and Validity Reliability; What you measure can also be measured by others in the same way? or how much of agreement is there between or among observers. Does this apply to your study?

Validity; Is the info which you got means what it is supposed to mean. Here in this case you have data from an open-access platform. Is this info True or only for open access?

original numerical values are used instead of percentages in comparisons or in associations as far as i know.

Kindly consider these issues in your analysis.

I am a single coder. I am doing coding twice at different points in time. Yes, this is open-access public data in reports. Bhogaraju Anand Which tests for reliability and validity are suitable in this case?

Hi,

As I understand, if there are different sources giving the same info about an organization then the issue of reliability arises. i.e how close or distant is the information. This is something like alternate form reliability. To know whether the info is true to its original statement or report something like an Audit report or some other such genuine report is needed. This may be not easy to acquire.If you have that info then you are looking for content validity. If you are looking for progress over time like performance then predictive validity. In case you are looking for groups and how much they distinguish between groups that it should theoretically be able to distinguish between i.e concurrent validity.

Hi Bhogaraju Anand Thanks for your help. Your suggestions seem working for me. I added up the scores of sub-categories to form the category column, and then added the categories score to form the Variable column. These variables are proxies of the broad concepts. Now, all the variables, categories and sub-categories have numeric value like DVs have the values as 0,0.33,0.67 or 1. Similar is with IVS and Mediating variables, having nominal values.

In this case, when I apply multinomial logistic regression, results are significant but in the parameter estimates table, what I see under the DV column are these values as 0, 0.33, 0.67 or 1. Does it look nice on the table? I was watching some videos and they use category labels but I do not have category labels in columns under rows. I have the composite score or percentage in categories instead. This means that IV is significant to report at least two categories(or 67%) DV out of the total three categories.

Also, as you suggested, can I add crosstabs as a preliminary analysis before Multinomial logistic regression. For crosstabs, should I do as rows, columns of these scored categories and sub-categories? Usually, in videos, they show labelling of categories, but I have scores.

And how do I find interactions between variables or categories? Maybe a silly question though!

To go deeper, as you suggested, I am now planning to consider all categories as variables and see the detailed analysis of each category in logistic regression. Let's see what comes out.

Hi,

If any variable has categories then one can label them. If values are on a continuous scale then parametric stats will work provided data is not skewed.

If feel always one should be able to do cross tabs with layering wherever necessary. It would also be nice if we can plot variables as it gives a visual impression and better understanding.

Nice to see you are progressing.

This is how I model. This i quite complicated to draw the model for mediation because it looks like SEM case due to many DVs and more than one mediators.

Do you think we can do mediation with logistc regression when we have more IVs, DVs and Mediators?

Main equation of IVs and DVs

DV1+DV2+DV3+DV4+DV5= IV1+IV2+ Control Var1+CV1+CV2+CV3+CV4+CV5

Mediation: Indirect effects

Mediation step1: M1+M2= IV1+IV2+ CV1+CV2+CV3+CV4+CV5

Mediation step2: DV1+DV2+DV3+DV4+DV5=M1+M2

Where

IV=independent variable

DV=Dependent variable

M=Mediator

CV=control variable

Bhogaraju Anand Moreover, if you think the model mentioned in the previous message is not workable, then I will think of summing up M1 and M2. Also, I will sumup all 5DVS as 1. The only big trouble are many DVs and mediators, but if we can then the analysis will be more detailed.