community
cancel
Showing results for 
Search instead for 
Did you mean: 

Alteryx designer Discussions

Find answers, ask questions, and share expertise about Alteryx Designer.

Correlation study on same data, but across multiple time ranges

Meteor

Hey everyone, thanks in advance for help on this!

 

I'm running Association Analysis on a number of different macroeconomic time series - commodity prices, inflation rate, etc. I have a complete data set of monthly data going back to 2000, but I set up filters + Association Analysis tool 4x up for the following time ranges:

1) 2000-onwards (full historical)

2) Previous 3 years (so 2016-2019)

3) Previous 5 years (so 2014-2019)

4) Previous 10 years (so 2009-2019)

 

I'm essentially wanting to compare how correlations are shifting over different frames. HOWEVER, when I run the analysis each of the four outputs of the Association Analysis tools are giving me identical correlations and p-values. Despite having different time ranges!

 

How is this possible? What am I missing?

 

Thanks again for the help,

Matt

Alteryx
Alteryx

Hi @mattd17 

 

Are you able to attach your workflow?

 

In the Association Analysis tool, are you referring to a targetted field? Even though the time ranges are different, the correlation and relationship of the data to your targetted field may be the same.

 

It also depends what fields you are selecting in the tool.

Meteor

Hi @RishiK,

 

There's some proprietary data mixed in with economic time series (Brent Oil, Sugar #11, USD FX rates) so I'm not sure I can post without losing some of the data needed for inference.

 


@RishiK wrote:

Hi @mattd17 

 

Are you able to attach your workflow?

 

In the Association Analysis tool, are you referring to a targetted field? Even though the time ranges are different, the correlation and relationship of the data to your targetted field may be the same.

 

It also depends what fields you are selecting in the tool.


In reference to the bolded/underlined section of your reply, I think this is highly unlikely. They should change at least a little bit just based on my fundmental understanding of economic forces changing over time. Additionally:

  • I ran checks in excel and the correlations also change with each time frame
  • I ran the same 4 different time ranges using just the Pearson Correlation tool - and I get different results for R2 in each matrix on each different time range

Why does the Pearson Correlation tool not generate the same results as choosing the Pearson Correlation measure in the Association Analysis?

 

As a follow-up, is it possible to get the p-values out of the Pearson Correlation tool results?

 

Thanks,

Matt

Alteryx
Alteryx

Hi @mattd17 

 

With regards to the differences between the tools, I will have to take a look at your workflow to see why you have different data reported out.

 

The attached macro can be used to get the report data from the data investigation and predictive tools

 

It was originally for the R output of the Linear Regression tool but you can tweak it for the Pearson Correlation and Association Analysis one I'd imagine.

Labels