Building models with SAS Enterprise Miner, SAS Factory Miner, SAS Visual Data Mining and Machine Learning or just with programming

Hypothesis testing with ASE?

Accepted Solution Solved
Reply
Occasional Contributor
Posts: 10
Accepted Solution

Hypothesis testing with ASE?

Hello,

 

I analyse texts with Text Mining in SAS EM and already did a decision tree plus a logistic regression.

As a result I am getting e.g. an average square error (ASE). Can this indicator sufficently accept a hypothesis?

Or are there better indicators?

Thanks, in advance and hope you have an idea.

 

Kind regards,

Benjamin


Accepted Solutions
Solution
‎01-25-2016 12:35 PM
SAS Employee
Posts: 29

Re: Hypothesis testing with ASE?

Posted in reply to Benjamin8

H Benjamin,

ASE is an estimate of model's mean squared error and it is not directly used for hypothesis testing. For example, in predictive modeling, you can use your model’s ASE on training and validation data to get an indication of overfitting. If your goal is overall hypothesis testing (between the target and all of the input variables), then you need to look at the overall p-value in the ANOVA table (available as on output of the Regression node). Total variance explained by the model and the model’s ASE (also degrees of freedoms of the model) are used together to calculate this p-value.

Hope this helps!

Funda

View solution in original post


All Replies
Super Contributor
Posts: 337

Re: Hypothesis testing with ASE?

Posted in reply to Benjamin8

Hi Benjamin,

More context please.

 

What does your Text Mining flow look like? And what are you trying to predict?

A general walk-through of your data and your goal would be nice too!

 

Thanks,

Occasional Contributor
Posts: 10

Re: Hypothesis testing with ASE?

Posted in reply to M_Maldonado

Hi Migual,

 

The Text Mining flow contains nothing special. In detail it contains Text Parsing-, Text Filter-, Text Topic-,Text Cluster-, Data Partition-, Decision Tree- and Regression-Node. I try to determine, if the text topics (independent var.) can explain a metrical targer variable (e.g. company size). A null hypothesis could be: There is no correlation between the text topics and the company size.

Hope this make things clearer.

Solution
‎01-25-2016 12:35 PM
SAS Employee
Posts: 29

Re: Hypothesis testing with ASE?

Posted in reply to Benjamin8

H Benjamin,

ASE is an estimate of model's mean squared error and it is not directly used for hypothesis testing. For example, in predictive modeling, you can use your model’s ASE on training and validation data to get an indication of overfitting. If your goal is overall hypothesis testing (between the target and all of the input variables), then you need to look at the overall p-value in the ANOVA table (available as on output of the Regression node). Total variance explained by the model and the model’s ASE (also degrees of freedoms of the model) are used together to calculate this p-value.

Hope this helps!

Funda

Occasional Contributor
Posts: 10

Re: Hypothesis testing with ASE?

Posted in reply to Funda_SAS

Hi Funda,

 

This sounds very good to me, the p-value is what I have been seeking for. I will try this.

Thank you.

 

Kind regards,

Benjamin

Occasional Contributor
Posts: 10

Re: Hypothesis testing with ASE?

Posted in reply to Benjamin8

Hi,

 

I tried this at SAS EM, but unfortunately I couldn't find the mentioned ANOVA table (with the p-value) of the Regresssion node.

Where can I find this table?

 

Best regards,

Benjamin

 

 

SAS Employee
Posts: 29

Re: Hypothesis testing with ASE?

Posted in reply to Benjamin8

ANOVA (Analysis of Variance) table is avalibale as an output of the regression node. P-value is shown by the pink arrowanova.png

Occasional Contributor
Posts: 10

Re: Hypothesis testing with ASE?

Posted in reply to Funda_SAS

OK, I was in the wrong node (decision tree) Smiley Wink  Now I've found it.

Thank's again.

☑ This topic is solved.

Need further help from the community? Please ask a new question.

Discussion stats
  • 7 replies
  • 580 views
  • 0 likes
  • 3 in conversation