linear regression on excel dataset

I am new in MATLAB and have tried to do a linear regression with the code:

>> filename = 'C:\Users\Troels\Dropbox\Analyse & Resultater\MATLAB\Danmark OX.xls';

>> ds = xlsread(filename) // the dataset "Danmark OX.xls" is printet

>> mdl = LinearModel.fit(ds)

Error using classreg.regr.TermsRegression/handleDataArgs (line 629) Y argument is required unless X is a dataset.

Error in LinearModel.fit (line 891) [X,y,haveDataset,otherArgs] = LinearModel.handleDataArgs(X,varargin{:});

Have also tried the code:

>> ds = dataset('XLSFile','C:\Users\Troels\Dropbox\Analyse & Resultater\MATLAB\Danmark OX.xls','ReadObsNames',true);

Warning: Variable names were modified to make them valid MATLAB identifiers.

> In @dataset\private\genvalidnames at 56

In @dataset\private\setvarnames at 40

In dataset.readXLSFile at 49

In dataset.dataset>dataset.dataset at 352

>> mdl = LinearModel.fit(ds);

Warning: Regression design matrix is rank deficient to within machine precision.

> In TermsRegression>TermsRegression.checkDesignRank at 98

In LinearModel.LinearModel>LinearModel.fit at 944

Am I on the right track in any of my 2 attempts? And can anyone tell me from the above what I do wrong?

[dpb]

Please reformat the code and remove the excess lines for legibility...

But, the first fails because the LinearModel requires a dataset and the result of xlsread() isn't one...

Second effort is ok from that standpoint but the error/warning indicates your data are poorly scaled -- lookfor "rank deficiency" in the online documentation for what that is if you don't know and possible workarounds.

Better would be to fix that problem w/ better design matrix but sometimes that's not possible.

Answer 1

[Shashank Prasanna]

mdl = LinearModel.fit(ds)

assumes, ds is a dataset (Your second approach) and the last column of ds is the response variable. Is the true in your case?

If you want to pass data as a matrix then you have to do so this way:

mdl = LinearModel.fit(X,y)

I urge you to read the documentation of LinearModel.fit to understand how to call it. This will save you a lot of time later on. There are plenty of examples there:

http://www.mathworks.com/help/stats/linearmodel.fit.html

Answer 2

[T27667]

Ohh, my second attempt actually worked and the estimates and so on were produced in spite of the warnings.

I miss some output however compared to the output from another statistical software (OxMetrics) which gives:

                   Coefficient  Std.Error  t-value  t-prob Part.R^2
Constant               1.36882     0.2155     6.35  0.0000   0.0708
Index                -0.936838    0.06085    -15.4  0.0000   0.3090
Institutionel        -0.515027    0.06092    -8.45  0.0000   0.1188
Risk                 0.0352812   0.004220     8.36  0.0000   0.1165
Equity                0.248302    0.05861     4.24  0.0000   0.0328
Allocation            0.375370    0.05182     7.24  0.0000   0.0901
Performance fee      0.0138219   0.007027     1.97  0.0497   0.0072
Share Class TNA     -0.0323253    0.01039    -3.11  0.0020   0.0179

sigma                 0.33081  RSS                58.0007371
R^2                  0.614132  F(7,530) =    120.5 [0.000]**
Adj.R^2              0.609035  log-likelihood       -164.218
no. of observations       538  no. of parameters           8
mean(Y)               1.22317  se(Y)                0.529066

Normality test:   Chi^2(2)  =   44.367 [0.0000]**
Hetero test:      F(10,527) =   2.5403 [0.0054]**
Hetero-X test:    F(13,524) =   2.0211 [0.0175]* 
RESET23 test:     F(2,528)  =   5.7023 [0.0035]**

How can I produce the various tests in MATLAB that OxMetrics provide above?

[Shashank Prasanna]

I've never used OxMetrics, but you may want to check the documentation of LinearModel.fit for all the calculated goodness of fit statistics. If you don't find a specific one it may be available separately in the Statistics Toolbox.