Subscribe Bookmark RSS Feed

Semi-quant vs. continuous data: best method to explore relationship?

sornasst

Community Trekker

Joined:

Feb 2, 2016

Hello again,

 

I need your advice about a rather basic question of statistics: what is the best method to explore the relationship (e.g. Spearman's correlation, Kendall's Tau, other) between semi-quantitative data (graded 0,1,2,3) and continuous variables (see table attached and screenshot)?

 

Mock Semi-Quant by Continuous.PNG

Thanks

1 ACCEPTED SOLUTION

Accepted Solutions
markbailey

Staff

Joined:

Jun 23, 2011

Solution

I would start with ordinal logistic regression. Assuming that you have the ordinal modeling type for the DATA SEMI-QUANT response and the continuous modeling type for the DATA CONTINUOUS predictor/factor, you can use either Analyze > Fit Y by X or Analyze > Fit Model. The Logistic platform (from Fit Y by X launch) provides initial results like this:

Untitled.jpeg

You can also get the ROC Curve and Lift Curve with this platform. Alternatively, you get these initial results from Ordinal Logistic Fit (from Fit Model launch):

Untitled.jpeg

This platform allows more effects in the model such as other predictors and covariates and their transforms (interactions, powers). The Prediction Profiler and the same diagnostic plots are also available.

These regression techniques provide more information than simple measures of association or agreement.

Learn it once, use it forever!
1 REPLY
markbailey

Staff

Joined:

Jun 23, 2011

Solution

I would start with ordinal logistic regression. Assuming that you have the ordinal modeling type for the DATA SEMI-QUANT response and the continuous modeling type for the DATA CONTINUOUS predictor/factor, you can use either Analyze > Fit Y by X or Analyze > Fit Model. The Logistic platform (from Fit Y by X launch) provides initial results like this:

Untitled.jpeg

You can also get the ROC Curve and Lift Curve with this platform. Alternatively, you get these initial results from Ordinal Logistic Fit (from Fit Model launch):

Untitled.jpeg

This platform allows more effects in the model such as other predictors and covariates and their transforms (interactions, powers). The Prediction Profiler and the same diagnostic plots are also available.

These regression techniques provide more information than simple measures of association or agreement.

Learn it once, use it forever!