Choose Language Hide Translation Bar
Highlighted
dkraeuter_sunne
Level III

Contribution of a single data point to the final model

I am trying to build a model with some manufacturing run data that I have. Is there a way to determine the amount one specific data point is influencing the final models created?

1 ACCEPTED SOLUTION

Accepted Solutions
Highlighted
julian
Community Manager Community Manager

Re: Contribution of a single data point to the final model

Hi @dkraeuter_sunne,

 

A common measure for this purpose is Cook's Distance, which is available for standard least squares analyses from Fit Model by going to the Red Triangle > Save Columns > Cook's D Influence. Doing this will return a new column to the data table with a Cook's distance for each row, which will allow you to investigate whether or not there are any rows with a concerning amount of influence on the model estimates. There are various guidelines for what counts as an overly influential point, but most agree a D > 1 is concerning. I find it helpful to look at the overall distribution of Cook's values by using Analyze > Distribution. 

Hope this helps!

 

@julian

View solution in original post

1 REPLY 1
Highlighted
julian
Community Manager Community Manager

Re: Contribution of a single data point to the final model

Hi @dkraeuter_sunne,

 

A common measure for this purpose is Cook's Distance, which is available for standard least squares analyses from Fit Model by going to the Red Triangle > Save Columns > Cook's D Influence. Doing this will return a new column to the data table with a Cook's distance for each row, which will allow you to investigate whether or not there are any rows with a concerning amount of influence on the model estimates. There are various guidelines for what counts as an overly influential point, but most agree a D > 1 is concerning. I find it helpful to look at the overall distribution of Cook's values by using Analyze > Distribution. 

Hope this helps!

 

@julian

View solution in original post

Article Labels

    There are no labels assigned to this post.