Slope / Sensitivity / Correlation

[matt23]

Hi, I need help getting something similar to this:

Here's a tiny bit of my data:

Year	Month	Day	Hour	Load	Temperature
2009	1	1	0	70232	20.87
2009	1	1	1	68422	20.61
2009	1	1	2	67014	20.27
2009	1	1	3	66068	20.52
2009	1	1	4	65781	21.45
2009	1	1	5	66308	21.69
2009	1	1	6	67559	21.87
2009	1	1	7	69150	22.39

Y = Load , X = Temp

[Reeza]

What are your definitions of slope and correlation here?

Do you have SAS/ETS licenced?

You can check with:

proc product_status;run;

---

@matt23 wrote:

Hi, I need help getting something similar to this:

Here's a tiny bit of my data:

Year	Month	Day	Hour	Load	Temperature
2009	1	1	0	70232	20.87
2009	1	1	1	68422	20.61
2009	1	1	2	67014	20.27
2009	1	1	3	66068	20.52
2009	1	1	4	65781	21.45
2009	1	1	5	66308	21.69
2009	1	1	6	67559	21.87
2009	1	1	7	69150	22.39

Y = Load , X = Temp

---

[matt23]

Actually working on 9.4 version right now

[pau13rown]

im not sure it makes any sense. why is slope not negative in the 20s where it's dipping down? correlation=autocorrelation?

[Reeza]

@pau13rown The graph is of slope over time, so the first image which is based off the second table below I assume. 

@matt23 What are your definitions of slope and correlation? SAS is packaged by modules, so 9.4 does not tell me if you have SAS /ETS or specifically if you can use PROC EXPAND which is used to calculate moving statistics easily. 

[pau13rown]

oh ok, load "sensibility"

[matt23]

proc product_status;run  

 For SAS/ETS ...
Custom version information: 14.2

Sorry, I don't know what it means

[Reeza]

Still need to know how you're defining slope and correlation here.

---

@matt23 wrote:

proc product_status;run  

 For SAS/ETS ...
Custom version information: 14.2

Sorry, I don't know what it means

---

[matt23]

Load = Intercept + Slope * Temperature

So I believe the slope is from regression at a specific time if I am not mistaken

Correlation is a correlation between load and temperature at time 

I don't know if it makes sense

[Reeza · Accepted solution]

You can't have a correlation or slope with only a single or even two observations, so it's a moving statistics of some kind. You need to define the window. 

 Depending on the window and exact calculations I'd probably recommend either an array approach or PROC EXPAND.

Here's the PROC EXPAND documentation:

http://documentation.sas.com/?docsetId=etsug&docsetTarget=etsug_expand_examples04.htm&docsetVersion=...

Available transformations:

http://documentation.sas.com/?docsetId=etsug&docsetTarget=etsug_expand_details19.htm&docsetVersion=1...

Otherwise, an array.

http://support.sas.com/kb/25/027.html

You'll likely need to write the formula for the slope/intercept either way, by determining the formulas/components and doing the calculation.