topic Re: WHAT DO THESE PLOTS SAY ABOUT THE REGRESSION OUTCOMES? in New SAS User

WHAT DO THESE PLOTS SAY ABOUT THE REGRESSION OUTCOMES?

d6k5d3 — Thu, 30 May 2019 05:34:18 GMT

I have run a regression, and in the output the following charts are generated. I cannot relate these charts with the parameter estimates. Could someone give some idea?

For example, does the chart imply that the predicted value of Abs_Jump is 0 by all the regressors except Abs_Surp_Mex14? How could this be if I have the parameter estimate for Abs_Surp_Mex8 which is 0.469 with p-value 0.01? Please enlighten me.

Re: WHAT DO THESE PLOTS SAY ABOUT THE REGRESSION OUTCOMES?

PaigeMiller — Thu, 30 May 2019 12:23:40 GMT

Can we see the code you used? Can we see the rest of the output from this regression?

Re: WHAT DO THESE PLOTS SAY ABOUT THE REGRESSION OUTCOMES?

Reeza — Thu, 30 May 2019 15:05:55 GMT

Why is your title in upper case?
Random guess based on not a whole lot, you have a rare event or data that's very weirdly distributed in some manner. What does the distribution of your response variable look like?

Re: WHAT DO THESE PLOTS SAY ABOUT THE REGRESSION OUTCOMES?

d6k5d3 — Thu, 30 May 2019 19:15:53 GMT

proc qlim data= MXN_Jump_News;
model Abs_Jump= DW_Mon DW_Tue DW_Wed DW_Thu Sin1 Sin2 Sin3 Sin4 Cos1 Cos2 Cos3 Cos4 Abs_Surp_US4
		Abs_Surp_US7 Abs_Surp_US8 Abs_Surp_US9 Abs_Surp_US11 Abs_Surp_US16 Abs_Surp_US20
		Abs_Surp_US22 PS_US30 Abs_Surp_GER12 Abs_Surp_MEX5 Abs_Surp_MEX6 Abs_Surp_MEX8
		Abs_Surp_MEX10 Abs_Surp_MEX14 PS_MEX16 SP4 SP16 SP21 SP23; nloptions maxiter= 500;
endogenous Abs_Jump ~ censored (lb= 0);
run;

DW_Mon to DW_Thu are dummies. SP4 SP16 SP21 SP23 are also dummies. Sin1-Sin4 and Cos1-Cos4 are sine and cosine values. The dependent variable has only positive values. The other independent variables take both positive and negative values.

Re: WHAT DO THESE PLOTS SAY ABOUT THE REGRESSION OUTCOMES?

d6k5d3 — Thu, 30 May 2019 19:22:34 GMT

Sorry for the upper case.

Re: WHAT DO THESE PLOTS SAY ABOUT THE REGRESSION OUTCOMES?

d6k5d3 — Thu, 30 May 2019 19:38:17 GMT

Would you please look at the code and the full result above? Much thanks.

Re: WHAT DO THESE PLOTS SAY ABOUT THE REGRESSION OUTCOMES?

PaigeMiller — Thu, 30 May 2019 20:18:55 GMT

As a general rule, when you get such un-intuitive results from a model fit, the causes are most likely one of the following

Extreme outlier(s)
Overfitting
Extreme multi-collinearity between the X-variables
Humongous amounts of noise in the y-variables — possibly because of an extreme outlier(s)

Now, I have never used PROC QLIM and so I can't really say any more about your results, and there may be other reasons why this is happening specifically to PROC QLIM. Nevertheless, the above 4 possibilities are things you need to investigate yourself, since you have the data.

Re: WHAT DO THESE PLOTS SAY ABOUT THE REGRESSION OUTCOMES?

ballardw — Thu, 30 May 2019 22:35:31 GMT

When I see code such as (reformatted for legibility on the forum)

proc qlim data= MXN_Jump_News;
model Abs_Jump= DW_Mon DW_Tue DW_Wed DW_Thu 
      Sin1 Sin2 Sin3 Sin4 Cos1 Cos2 Cos3 Cos4 
      Abs_Surp_US4 Abs_Surp_US7 Abs_Surp_US8 Abs_Surp_US9 
      Abs_Surp_US11 Abs_Surp_US16 Abs_Surp_US20 Abs_Surp_US22 PS_US30 
      Abs_Surp_GER12 
      Abs_Surp_MEX5 Abs_Surp_MEX6 Abs_Surp_MEX8 Abs_Surp_MEX10 Abs_Surp_MEX14 PS_MEX16 
      SP4 SP16 SP21 SP23; nloptions maxiter= 500;
endogenous Abs_Jump ~ censored (lb= 0);
run;

I wonder if when the values for the variables with US in the name are populated whether the ones with GER or MEX have values other than 0 and vice versa. Maybe just too many variables with 0 for values.

It may not hurt to describe, if not list some data, why the Abs_Surp variables have such different suffixes.

Re: WHAT DO THESE PLOTS SAY ABOUT THE REGRESSION OUTCOMES?

Reeza — Fri, 31 May 2019 16:16:17 GMT

@d6k5d3 wrote:
Would you please look at the code and the full result above? Much thanks.

My previous question is still unanswered, which will likely help answer the other questions. What does the distribution of your response variable look like before modelling? If you take a close look at your output it seems like there isn't even an upper bound which seems a bit weird to me, which may mean your data is massively skewed as I suspected.