don't understand output of fitlme, intercept substituted for first line.

Hi - I fit some linear mixed effect models using fitlme , but the output always substitutes 'intercept' for the first line of the fixed effect coefficiants. Why is this?

Fixed effects coefficients (95% CIs):

Name Estimate SE tStat DF pValue Lower

{'(Intercept)' } -1.984e-05 6.4305e-06 -3.0853 37 0.0038374 -3.2869e-05

{'voc_registers_ADS' } -9.3714e-06 6.9456e-06 -1.3493 37 0.18545 -2.3444e-05

{'voc_registers_nonSpeech'} 5.4826e-06 6.4513e-06 0.84985 37 0.40088

Answer 1

[the cyclist]

The output is not substituting 'Intercept' for a level of your coefficient.

It looks like your variable voc_registers is categorical, and has three categories:

• ADS
• nonSpeech
• anotherOne <------ Not possible for me to know the name, from the output

In this case, MATLAB is using 'anotherOne' as the reference category. (This is described in the documentation for fitlme.) By construction, the coefficient of the reference category is zero, and it is not listed in the output.

'Intercept' is the estimate of the mean when continuous explanatory variables are zero, and categorical explanatory variables are at the reference. So, if your model is

'y ~ alpha + beta*voc_registers'

it is the estimate of the value of alpha.

It can be a bit more complicated, if you have both fixed and random effects for the intercept.

[Emu]

Hi sorry for the delay I got really sick.

[Emu]

Ok so, whichever category is used as the reference will always be zero - so if I wanted to know the coefficients of this category, I could just input the categories in a different order? (does it work like that?)

I have both random and fixed effects.

[the cyclist]

Sorry to hear you were sick. I hope you are doing better.

Yes, you could rearrange the categories to make a different one the reference. (I think it chooses the category of the first observation as the reference? I'm not sure. The documentation will probably say.)

But, you won't learn any new information. The coefficients are always relative to the other categories. If you want voc_registers_ADS to be the reference, just add 9.3714e-06 to the other two coefficients.