DATA AND STATISTICS - Georgia State University

advertisement
MGS 8150
Causal Model – extra
Dr. Subhashish (Sub) Samaddar
Georgia State University
J. Mack Robinson College of Business
Executive Education
Atlanta, GA 30303
Slide 1
Causal Model: Some Useful Tips




Choose and reason your Dependent Variable Y and
Independent variable (X) carefully. Be able to reason:
A change in X should cause a change in Y AND a
change in Y should not cause a change in X.
Your data for Y should have variance – no variance is
bad.
Your data for each X variable should have variance –
no variance is bad.
Recognize variable types that you are dealing with
and take appropriate action:
• Four
• Nominal
• Ordinal
• Interval/ Ratio
Slide 2
Causal Model: Role of Variable Types



Not all variables created equal! Based on
amount of information contained in the data
(or variable)
Why do we care – to be able to use them
appropriately in causal modeling
How many different types of variables?
• Four
•Nominal
•Ordinal
•Interval/ Ratio
Slide 3
Scales of Measurement


Nominal – data contains only name or label to
describe an attribute; can be numeric or non-numeric.
 Example:
 University students data can use a
nonnumeric label such as Business,
Humanities, Education, and so on.
 Gender – male/ female.
How to model this type of data:
 Use dummy variable; easy for two values such as
Gender – Use dummy variable X1 where X1 = 0
means female, X1 = 1 means male. If you have
more than two values talk to Sub.
Slide 4
Scales of Measurement
Ordinal – nominal data properties plus there is a
meaningful order or rank of the data; can be
numeric or non-numeric
Examples:
1. University students data can use a nonnumeric
label such as Freshman, Sophomore, Junior,
or Senior.
2. Military ranks …

How to model this type of data:
These can use numeric code … such as 1, 2, 3, 4 etc.
where 2 represents something more than 1 and so
on.
Slide 5
Scales of Measurement
Interval – ordinal data properties plus a fixed unit of
measure expressing the interval between the
observations; always numeric.
Ratio – interval data properties plus ratio of two values
are meaningful.
Examples:
1. (Interval data) John’s exam score is 87, Jane’s score is
94. Jane scored 7 points more than John.
2. (Ratio data) Distance, Height, Weight, Time, Money
…
How to use them in causal model: The regression
model that you can run in Excel has to have an
Interval or ratio data (variable) as the dependent
variable. X variables can be any type.
Slide 6
How much data do you need?
Some rule of thumbs:
1. Keeping it simple, it depends on how many
X variables you have in your model.
2. Will discuss some rule-of-thumb in class:
Use the largest of:
a. 50+8*k (for R-squared test only)
b. 104+k (for coefficients tests only)
Where k represents number of X variables in
your regression model.
Slide 7
Slide 8
Download