MixedLM Predict issue

[prashanthin]

Hi,

I am doing some research on Mixed effects models and using Stats Mixedlm for same. I have Units to be predicted from Price and size at different UPC levels.

So below is the code I used -

md = smf.mixedlm("Units ~ Avg_Units_Price + C_CUSTOM_BASE_SIZE", file3, groups=file3["UPC_10_digi"], re_formula="~Avg_Units_Price + C_CUSTOM_BASE_SIZE")
mdf = md.fit()
print(mdf.summary())

Output

mdf.random_effects

I am getting moderate avp fit on predicted values however I have below needs -

1. I want to predict Units using both fixed and random effects as in below case -

• so I want to predict units at size 20 and Price 11.1. I know for predicting these values both fixed effects and random effects would be used, however I am not able to use predict function as I couldnt get clear example/idea for this.
  I have used below code for prediction but its showing multiple arguments error -

Can you please help me in this single case prediction using Mixedlm?

Thanks a lot in advance.

[kshedden]

There is no convenience method for this. You would need to take the BLUPs (from the random_effects attribute as above) and add them to the fitted values from the fixed effects (which is what you get from 'predict'). For example, for all observations in group 7023001154 you would take the fixed effects prediction and add to it -327.6 + 66*Avg_Units_Price - 72*C_CUSTOM_BASE_SIZE.

[snowoody]

One way to get the random effect is through the predict function in Linear Mixed Effects Model md.

md.predict(mdf.random_effects[group_a1_df_2[0]], exog=a_df)
would get you the random effect prediction.

model.predict(mdf.fe_params, exog=a_df)
would get you the fixed effect prediction.

You probably still need to add them together after the calculation. Got this from https://stackoverflow.com/questions/50300224/statsmodels-mixed-linear-model-predictions. It works for me. But I have to do some preprocessing for exog as it does not take my pandas dataframe directly for categorical values.