Regression problem: How we can Ignore Outliers in a regression analyses??

alisr

New Member
#1
[Minitab 16]Regression problem: How we can Ignore Outliers in a regression analyses??

Hello
I want to draw a simple linear regression chart in Minitab and I want to find The regression equation. I use two column of data and this path:
Stat>>Regression>>Regression...
Minitab shows me outliers in the results but it doesn't give me the correct equation!!
I want to ignore these Outliers without deleting rows.
so what I can do about this??
Here is a picture of my question:

Thanks a lot
 
Last edited:

Miner

TS Contributor
#2
Could you explain the issue in greater detail? Your picture was blocked by the firewall. I am a Minitab user.
 

alisr

New Member
#3
Could you explain the issue in greater detail? Your picture was blocked by the firewall. I am a Minitab user.
Thanks for your answer.............
I uploaded image again:


I want to find a simple regression equation (like y=ax+b). so I used my data in two columns. but my data has Outliers. so I want to Ignore all of them. but I don't want to delete the rows, because that is difficult and I want to draw about 30 same diagrams.
for example, Imagine your data is:
y x
1 1
2 2
3 3
4 4
100 5

if you write these numbers in two columns and use Regression from this path:
stat>>Regression>>Regression..
and choose y as response and x as predictor and if you mark Result that shows the equation, so we have this in Results:



=====================================================
also and if we write these numbers:
y x
1 1
2 2
3 3
4 4
5 100
we have this:


============================================================
you know both of the answers are not correct because the point (100 5) and (5 100) are Outliers and correct equation is y=x for two answers.
also Minitab just shows Outlier in second answer but doesn't Ignore it.
My questions are:
1- how we can Ignore Outliers to have correct regression equation??
2- why minitab doesn't Recognize Outlier in first data??
Thanks a lot!
 
Last edited:

Miner

TS Contributor
#4
This is more complicated than just simply removing or ignoring outliers. These do not appear to be spurious results or data entry. It appears that there may be a lurking variable that changes the slope of the regression line.
 

alisr

New Member
#5
thanks
I draw a diagram between "Price" and "one of the indexes of Technical Analysis(that's name is RSI)" in Financial markets.
I sure that I can delete some of the wrong data with an algorithm in Excel because they don't have an important condition.
so as you said I think it is better to correct my data and then if I see outliers it has a meaning.

Do you agree with this??: because of that I think Minitab doesn't let us to Ignore Outliers!!
by the way:
1- why minitab doesn't Recognize Outlier in first data??
 
Last edited:

Miner

TS Contributor
#6
It does recognize the outliers, and unless the results are suppressed, it prints a list of unusual observations in the session window. These should have been flagged as having high leverage. Also, if you enable the "Four in one" residuals plot in the Graphs submenu, it would show them graphically.

The easiest way to remove them is to brush these points in the scatter plot or the residuals plot using the brushing tool then go to Data > Subset Worksheet > Specify which rows to exclude > Brushed rows.
 

alisr

New Member
#7
Thanks a lot for explaining about brush tools.
but I cant understand this matter about data:
as I said when
y x
1 1
2 2
3 3
4 4
100 5
minitab doesn't show unusual observations!!!why??
 

noetsi

Fortran must die
#8
Normally you would not remove outliers (which has nothing to do with Minitab, it is inherent in regression).

All software have rules for what is an outlier. For example the standardized residual might be greater than 3 before it is reported. If minitab is not reporting it as an outlier, then it did not break the rules it uses for that. You would have to look at the documentation for how it reports unusual data and then figure out why that point did not violate that rule. Say the rule was a standardized residual of 3 or more. You would look at a plot of standardized residuals, find that point, and see what the value is.