Calculating the same coefficients using same data in Matlab and SPSS

I'm trying to obtain the same coefficients from Matlab and SPSS using the same data set from an SPSS text book to determine whether my code is correct in Matlab.

My predictor varibables are as follows: x1 = advertising budget x2 = airplay x3 = aesthetics

My outcome variable is: y = record sales

The equation in SPSS uses the following equation: y = (b0 + b1x1 + b2x2 + b3x3) + error

where b0 and b1, b2, b3 are the regression coefficients the following coefficients were obtained in SPSS:

b1 = 0.578 b2 = 0.599 b3 = 0.326

Using the following Matlab code:

X = [ones(length(x1),1) x1 x2 x3];

I obtain the following coefficients from the same data set using B = X\y;

20.7809 0.0862 3.309 10.2938

which as you can see is considerably different.

Any help would be appreciated.

[Star Strider]

The data do not seem to be available online (they may be proprietary to SPSS), so it's impossible to reproduce your regression.

[Tim Bennett]

Thanks for getting back to me. I can send the data to you via email if thats an option? The link to the data i'm using is www.sagepub.co.uk/field3e (after selecting area location on the webpage, select "All SPSS Datafiles" then "Datafiles" and the data is saved as "record2.sav") but like you say the data is probably protected without an SPSS license.

Thanks

[Tim Bennett]

PS My email is [e-mail] if you're happy for me to send the data through to you?

[Star Strider]

It's a .sav file. I can open it but it's encoded in a way I can't figure out, . Maybe you could post the headers and perhaps about 15 lines of data here, or choose a different data file.

In the interim, I suggest you look at how the data files are organised to see if you're presenting the same data to both SPSS and MATLAB in the way that MATLAB expects.

[Tim Bennett]

I'll send each column of data (the top 10 data points each)through for each variable if thats o.k to keep things simple. The x (predictor variables) and y (outcome variable) data for both SPSS and Matlab are in column vectors [10 x 1]. x1 = advertising budget x2 = airplay x3 = band attractiveness

My outcome variable is: y = record sales

Using the following Matlab code:

X = [ones(length(x1),1) x1 x2 x3];

I obtain the following coefficients from the data set sent though using B = X\y;

-67.9982, 0.0466, 4.6682, 17.8968,

in a [4 x 1] vector

I'll send through the coefficients from SPSS for this data set asap.

Thanks

[Tim Bennett]

outcome data (y):

270, 220, 170, 70, 210, 200, 300, 290, 70, 150,

x1 data:

1188.19, 574.51, 568.95, 471.81, 537.35, 514.07, 174.09, 1720.81, 611.48, 251.19,

x2 data:

33, 44, 19, 20, 22, 21, 40, 32, 20, 24,

x3 data

7, 5, 5, 1, 9, 7, 7, 7, 2, 8,

[Tim Bennett]

Just obtained the same coefficient outputs from both methods so the matlab code works fine. Thanks for your help anyway Star Strider.

Answer 1

[Wayne King]

I'm not sure what is going on in SPSS because I don't use that, but I can tell you that repeating the above in R and MATLAB gives the same coefficients.

Here is the output I get in R:

Call:
lm(formula = y ~ x1 + x2 + x3)

Coefficients:
(Intercept)           x1           x2           x3  
  -68.00038      0.04661      4.66820     17.89690

Here is the coefficient vector I get in MATLAB:

beta = [ -67.9982
           0.0466
           4.6682
          17.8968]

[Tim Bennett]

Yes, i've just obtained the same coefficient outputs from both methods so the matlab code works fine. Thanks for your help.