How to calculate mean error(ME) in sas

[Cecillia_Mao]

Hi everyone, does anyone know how to calculate mean error(ME) in SAS. I checked the proc means, its statistical list does not contain ME. So I'm wondering are there any procedures that can calculate ME directly? Thanks for any hints.

[PaigeMiller]

In a data step, compute the value of y minus y-hat. Then in PROC MEANS or PROC SUMMARY, compute the mean of this difference.

If that's not what you are looking for, then a few more details would help. What is meant in the formula by y-hat? Does it have to do with predicting (the word "prediction" appears) from a model? If so, what type of model (and what SAS PROC) creates this model? What are the variable names in your SAS data set? 

[Cecillia_Mao]

Thanks a lot! I really appreciate it! Sometimes, if there is no available options, we do need to calculated step by steps.

[ballardw]

If y sub i is an individual value and the y hat is the predicted then you are looking at residuals and that looks like the mean of residuals. If so, then you run a regression and get the residuals in the output. Then take the mean of the residual.

[PaigeMiller]

---

@ballardw wrote:

If y sub i is an individual value and the y hat is the predicted then you are looking at residuals and that looks like the mean of residuals. If so, then you run a regression and get the residuals in the output. Then take the mean of the residual.

---

Which is zero for many types of models. No need to compute it (if this comes from one of those types of models)

[ballardw]

---

@PaigeMiller wrote:

---

@ballardw wrote:

If y sub i is an individual value and the y hat is the predicted then you are looking at residuals and that looks like the mean of residuals. If so, then you run a regression and get the residuals in the output. Then take the mean of the residual.

---

Which is zero for many types of models. No need to compute it (if this comes from one of those types of models)

---

Yep. Which is partially why I was asking what the y values in the formula represent.

[Cecillia_Mao]

Thank you Ballardw, I'm sorry for the late reply. I was confused with the following two staitistics: "mean error" and "mean deviation" mean deviation,which is sum(X-X̄)/n. Thanks for clarification.

[Reeza]

For the Mean error you need an observed/recorded value minus the predicted value, which comes from a model. 

If you only have a single value, the typical estimate ends up being the mean, so then your formula changes to 

1/n*sum(observed - mean(x))

which has a value of 0 always. I'll leave the algebra as an exercise for you, or for you to google at least 😉

Unless you have a modeled value, this is a 'trick' question. This is why the mean squared error or the mean absolute error measures are used instead.

---

@Cecillia_Mao wrote:

Hi everyone, does anyone know how to calculate mean error(ME) in SAS. I checked the proc means, its statistical list does not contain ME. So I'm wondering are there any procedures that can calculate ME directly? Thanks for any hints.

---

[Cecillia_Mao]

Thank you so much Reeza. That totally make sense why mean squared error or the mean absolute error measures are used more often. I was supposed to ask how to calculate mean deviation. Unlike to standard deviation, I believe meand deviation can't be calculated using proc means options directly.

[Ksharp]

If you have two columns Y and Y_Hat:

proc sql;
select mean(Y-Y_Hat) from have;
quit;