Comparing two variables for correlation and determining an equation to relate the two variable

[#1]

I think this is a linear regression method but I've forgotten how to perform the analysis.  I have two variables:

ACWP_VNDR_HW by Month_Yr 

AND

MAT by Month_Yr looking at the attached it looks like there could be some correlation.

I've attached some screen shots.  Can someone please advise if I'm on the right track or if not recommendations to proceed.  I'd like to develop an equation for VNDR_HR as a function of MAT.

Thanks!

[#2 · Accepted solution]

Not sure why you didn't just post up your data...but you are performing linear regression Fit Y by X.  The output of this analysis shows there is almost no relationship between the 2 variables (RSquare is ~0 and p value is .44).  If you want to get Pearson's correlation coefficient you could do Multivariate Methods>Multivariate.

If the variables do not correlate (and you think they should), here are some things to think about:

1. If either variable does not vary much in the data set, it is difficult to see relationships,

2. If there is a lagged effect, the data set would have to be modified to see this,

3. Unusual data points can make it difficult to see relationships (but there is little evidence in the data you present)

4. Your model may not contain the correct order of effects (missing higher order terms, but your data does not indicate this),

5. There are hidden variables not recorded in the data table,

6. You have measurement system issues.,

7. There is insufficient data to support  the detecting of a relationship.

Or there really is no relationship.

[#3]

Thank you for this information, it was very helpful

[#4 · Accepted solution]

@dfusco923 The only thing I can add to everything that @statman has contributed (with whom I agree on all points) is unless I'm missing something it looks like you might have a time series oriented dependent variable?

I'm basing the above question on the axis labels of the Graph Builder report. Maybe monthly aggregated data? If that's the case then using the Fit Y by X platform to establish some sort of correlation is misguided at best. If indeed you have a time series dependent variable, then some sort of time series modeling/analysis approach is recommended. A simple run chart and using what I call the Peter Bartell '3 second rule' is about all you need. My '3 second rule' says, "Stare at a plot for no longer than 3 seconds. Then make your decision or take action. Because the longer you stare at the plot, the more likely you'll talk yourself into seeing something that just isn't there." You can get more elegant by analyzing the data with a control chart...but the answer will be unchanged.

If I apply the 3 second rule to the Graph Builder report...there is nothing there. 

[#5]

Thank you for the additional information and recommendations they were very useful.