Hi Guys,
I am new to the list so I am going to quickly introduce myself:
I don't have any highly qualified degree in statistics, however because of
my job I need to work with statistics on regular bases.
I am currently being challenged by the following:
I have a sample of 500 people rating 3 questions (A, B & C) on a 0 to 10
scale, those 3 questions are then process through a linear equation of the
following type:
Yreg= a.B+b.B+c.C (a,b & c being steady coefficient)
The linear equation above is the output of a multiple linear regression
between Y which follows a normal law and A,B & C.
Is that OK to assume that Yreg will be likely to follow normal law ???
I need to know this as I want to apply some t-test to test for true
differences at a population level between an average based on Y reg and a
defined value.
My next question is even more challenging:
Now I cut my sample in 3 independent sub samples and I using 3 DIFFERENT
linear regressions for every of the sub samples based on the same variables
as used before (A,B,C & Y).
I get
Yreg1=a1A+b1B+c1C
Yreg2=a2A+b2B+c2C
Yreg3=a3A+b3B+c3C
I then compute a new variable based on a linear equation which coefficients
(J,K & L)have been defined arbitrary:
F=(J.Yavr1)+(K.Yavr2)+(L.Yavr3).
where Yavr1 is the output of averaging the Yreg1 output (i.e. if 10
respondents receive 10 Yreg1 scores I just divide the sumation of the 10
Yreg1 scores by10).
Is F likely to F follow a normal law & can I consequently use some t-test to
test for true differences at a population level between an average based on
F and a defined value ???
As the data haven't been collected yet I can't check for normality
Is there anyone on this planet that is able to help me ?
Olivier Van Parys
Senior Research Manager @ Amárach Consulting
28 Upper Fitzwilliam Street
Dublin 2
Tel: 01 6619147 Fax: 01 6610312
E-mail: [log in to unmask]
Web: www.amarach.com
|