Why does feval produce the error 'cannot automatically convert a double variable to categorical values'?

Hi,

I'm trying to fit a linear regression with a categorical and a continuous variable as dependent variables, and a continuous variable as the response. I then wanted to plot those on a scatter, with the regression lines drawn on. I was following the example given on this page: https://uk.mathworks.com/help/stats/regression-with-categorical-covariates.html

While that code works, mine doesn't seem to - I can create the model, plot the scatter, but when it comes to plotting the fitted regression lines, I get several errors, the first being 'cannot automatically convert a double variable to categorical values'.

Thanks in advance for your help!

This is my code:

%create the model
tbl=array2table(sctrdata(:,2:4),'VariableNames',{'time','proms','stat'})
tbl.time=categorical(tbl.time)
lm=fitlm(tbl,'proms~stat+time')

%plot the data
w = linspace(min(sctrdata(:,4)),max(sctrdata(:,4)));
figure()
gscatter(sctrdata(:,4),sctrdata(:,3),sctrdata(:,2),'bgr','x.o')
line(w,feval(lm,w,'1'),'Color','b','LineWidth',2)
line(w,feval(lm,w,'2'),'Color','g','LineWidth',2)
line(w,feval(lm,w,'3'),'Color','r','LineWidth',2)

This is the 18x4 double sctrdata:

202 1 35 7.35290974607409

204 1 15 16.1392276652092

207 1 29 8.61536130247136

210 1 30 18.6708615548317

213 1 16 18.6747515545782

214 1 24 7.42659245084273

202 2 30 5.65164424693076

204 2 10 9.95657468999142

210 2 32 15.3576732431412

213 2 35 7.39980077321328

207 2 NaN NaN

214 2 NaN NaN

204 3 25 12.1170053044237

210 3 43 4.81401465834751

213 3 46 11.3542368593944

202 3 NaN NaN

207 3 NaN NaN

214 3 NaN NaN

[Walter Roberson]

sctrdata = [
202	1	35	7.35290974607409
204	1	15	16.1392276652092
207	1	29	8.61536130247136
210	1	30	18.6708615548317
213	1	16	18.6747515545782
214	1	24	7.42659245084273
202	2	30	5.65164424693076
204	2	10	9.95657468999142
210	2	32	15.3576732431412
213	2	35	7.39980077321328
207	2	NaN	NaN
214	2	NaN	NaN
204	3	25	12.1170053044237
210	3	43	4.81401465834751
213	3	46	11.3542368593944
202	3	NaN	NaN
207	3	NaN	NaN
214	3	NaN	NaN];

%create the model
tbl=array2table(sctrdata(:,2:4),'VariableNames',{'time','proms','stat'})

tbl = 18x3 table
    time    proms     stat 
    ____    _____    ______

     1        35     7.3529
     1        15     16.139
     1        29     8.6154
     1        30     18.671
     1        16     18.675
     1        24     7.4266
     2        30     5.6516
     2        10     9.9566
     2        32     15.358
     2        35     7.3998
     2       NaN        NaN
     2       NaN        NaN
     3        25     12.117
     3        43      4.814
     3        46     11.354
     3       NaN        NaN

tbl.time=categorical(tbl.time)

tbl = 18x3 table
    time    proms     stat 
    ____    _____    ______

     1        35     7.3529
     1        15     16.139
     1        29     8.6154
     1        30     18.671
     1        16     18.675
     1        24     7.4266
     2        30     5.6516
     2        10     9.9566
     2        32     15.358
     2        35     7.3998
     2       NaN        NaN
     2       NaN        NaN
     3        25     12.117
     3        43      4.814
     3        46     11.354
     3       NaN        NaN

lm=fitlm(tbl,'proms~stat+time')

lm = 
Linear regression model:
    proms ~ 1 + time + stat

Estimated Coefficients:
                   Estimate      SE         tStat       pValue  
                   ________    _______    _________    _________

    (Intercept)       33.94     8.9073       3.8104    0.0041509
    time_2         -0.37315     6.5696    -0.056798      0.95595
    time_3           10.761     7.1697       1.5009      0.16762
    stat           -0.71071    0.62282      -1.1411      0.28327


Number of observations: 13, Error degrees of freedom: 9
Root Mean Squared Error: 9.69
R-squared: 0.365,  Adjusted R-Squared: 0.153
F-statistic vs. constant model: 1.72, p-value = 0.231

%plot the data
w = linspace(min(sctrdata(:,4)),max(sctrdata(:,4)));
figure()
gscatter(sctrdata(:,4),sctrdata(:,3),sctrdata(:,2),'bgr','x.o')

line(w,feval(lm,w,'1'),'Color','b','LineWidth',2)

Error using classreg.regr.modelutils.designmatrix>convertVar (line 505)
Cannot automatically convert a double variable to categorical values.

Error in classreg.regr.modelutils.designmatrix>dummyVars (line 437)
group = convertVar(group,glevels); % convert group to be compatible with glevels

Error in classreg.regr.modelutils.designmatrix (line 281)
[Xj,dummynames] = dummyVars(dummyCoding{j},Xj,catLevels{j});

Error in classreg.regr.CompactTermsRegression/designMatrix (line 142)
= classreg.regr.modelutils.designmatrix(X,'Model',terms(:,varLocs), ...

Error in LinearModel/predict (line 365)
design = designMatrix(model,Xpred);

Error in classreg.regr.CompactPredictor/feval (line 110)
yPred = reshape(predict(model,Xpred),sizeOut);

line(w,feval(lm,w,'2'),'Color','g','LineWidth',2)
line(w,feval(lm,w,'3'),'Color','r','LineWidth',2)

Answer 1

[Walter Roberson]

You design your tbl so that the first column is categorical.

You construct a fitlm model from that.

You attempt to feval() the model passing in a numeric value first and a character vector second. The numeric value is constructed from values based on column 4 of the original data.

Numeric values do not match the datatype of the first model parameter, which is categorical.

You should try

line(w,feval(lm,'1', w),'Color','b','LineWidth',2)

[Walter Roberson]

sctrdata = [
202	1	35	7.35290974607409
204	1	15	16.1392276652092
207	1	29	8.61536130247136
210	1	30	18.6708615548317
213	1	16	18.6747515545782
214	1	24	7.42659245084273
202	2	30	5.65164424693076
204	2	10	9.95657468999142
210	2	32	15.3576732431412
213	2	35	7.39980077321328
207	2	NaN	NaN
214	2	NaN	NaN
204	3	25	12.1170053044237
210	3	43	4.81401465834751
213	3	46	11.3542368593944
202	3	NaN	NaN
207	3	NaN	NaN
214	3	NaN	NaN];

%create the model
tbl=array2table(sctrdata(:,2:4),'VariableNames',{'time','proms','stat'})

tbl = 18x3 table
    time    proms     stat 
    ____    _____    ______

     1        35     7.3529
     1        15     16.139
     1        29     8.6154
     1        30     18.671
     1        16     18.675
     1        24     7.4266
     2        30     5.6516
     2        10     9.9566
     2        32     15.358
     2        35     7.3998
     2       NaN        NaN
     2       NaN        NaN
     3        25     12.117
     3        43      4.814
     3        46     11.354
     3       NaN        NaN

tbl.time=categorical(tbl.time)

tbl = 18x3 table
    time    proms     stat 
    ____    _____    ______

     1        35     7.3529
     1        15     16.139
     1        29     8.6154
     1        30     18.671
     1        16     18.675
     1        24     7.4266
     2        30     5.6516
     2        10     9.9566
     2        32     15.358
     2        35     7.3998
     2       NaN        NaN
     2       NaN        NaN
     3        25     12.117
     3        43      4.814
     3        46     11.354
     3       NaN        NaN

lm=fitlm(tbl,'proms~stat+time')

lm = 
Linear regression model:
    proms ~ 1 + time + stat

Estimated Coefficients:
                   Estimate      SE         tStat       pValue  
                   ________    _______    _________    _________

    (Intercept)       33.94     8.9073       3.8104    0.0041509
    time_2         -0.37315     6.5696    -0.056798      0.95595
    time_3           10.761     7.1697       1.5009      0.16762
    stat           -0.71071    0.62282      -1.1411      0.28327


Number of observations: 13, Error degrees of freedom: 9
Root Mean Squared Error: 9.69
R-squared: 0.365,  Adjusted R-Squared: 0.153
F-statistic vs. constant model: 1.72, p-value = 0.231

%plot the data
w = linspace(min(sctrdata(:,4)),max(sctrdata(:,4)));
figure()
gscatter(sctrdata(:,4),sctrdata(:,3),sctrdata(:,2),'bgr','x.o')
line(w,feval(lm,'1',w),'Color','b','LineWidth',2)
line(w,feval(lm,'2',w),'Color','g','LineWidth',2)
line(w,feval(lm,'3',w),'Color','r','LineWidth',2)

[David Bruce]

Ah, I see now. Thank you so much Walter and for responding so quickly!