Model looks good but outcome isn't stable

[devank]

Hi,
I'm hoping I can get some expert advice.

I have created a Linear Regression Model and I thought my results were good.
Model 1: Without WkDyPortion
Prediction was 7.3

Then I added another variable.
Model 2: With WkDyPortion
Prediction was 6

Why are they so different?
Which one should I trust more?
How would I know I'm not missing any other key independent?
Please help or point me in the right direction

Regards,
devank

[nirgrahamuk]

When you say prediction was 7.3
Prediction was 6
Do you mean that for one specific observation you have in mind, the first model predicted 7.3 for RestNB and the other 6?

What is the typical range of values of RestNB?

[devank]

I modelled last four years of March data. Predicted sales coming from remaining March days (eg: 13th -31st of Mar). Used observations from past March days (eg: 1st-12th Dec).

Each year we got between $5-$8K sales in March.
Sample size is about 600.
There are five years, so there are 120 subjects.

[nirgrahamuk]

This is interesting background, but can you specifically address my questions?
I.e. what does prediction was 7.3 mean in the context of what you said?

[FJCC]

I see that the median residual is on the order of -1000. If the predicted values of 7 and 6.3 are on the same scale as the residuals, then that difference seems very small.

[devank]

Thanks for looking at it nirgrahamuk and FJCC.

Yesterday's 7.3 prediction is now 7.48 today. It is 7,483,107 cents.
Apologies. I tried simplyfy problem but that may have caused more harm.

In above summary table,
Sum_X = Predicted value for rest of March. Actual prediction for 2020 March is 4,708,408
Sum_RestNB = Observded values from 14th - 31st of Mar.
Sum_ToDateNB= Observed values from 1st - 14th Mar.
accuracy = My simple calc to see how good the prediction is.
API = Sum_ToDateNB + Sum_X = 2,774,699 + 4,708,408 = 7,483,107