Fitting models using R-style formulas¶
Since version 0.5.0, statsmodels allows users to fit statistical
models using R-style formulas. Internally, statsmodels uses the
patsy package to convert formulas and
data to the matrices that are used in model fitting. The formula
framework is quite powerful; this tutorial only scratches the surface. A
full description of the formula language can be found in the patsy
docs:
Loading modules and functions¶
In [1]: import statsmodels.formula.api as smf
In [2]: import numpy as np
In [3]: import pandas
Notice that we called statsmodels.formula.api instead of the usual
statsmodels.api. The formula.api hosts many of the same
functions found in api (e.g. OLS, GLM), but it also holds lower case
counterparts for most of these models. In general, lower case models
accept formula and df arguments, whereas upper case ones take
endog and exog design matrices. formula accepts a string
which describes the model in terms of a patsy formula. df takes
a pandas data frame.
dir(smf) will print a list of available models.
Formula-compatible models have the following generic call signature:
(formula, data, subset=None, *args, **kwargs)
OLS regression using formulas¶
To begin, we fit the linear model described on the Getting Started page. Download the data, subset columns, and list-wise delete to remove missing observations:
In [4]: df = sm.datasets.get_rdataset("Guerry", "HistData").data
In [5]: df = df[['Lottery', 'Literacy', 'Wealth', 'Region']].dropna()
In [6]: df.head()
Out[6]:
Lottery Literacy Wealth Region
0 41 37 73 E
1 38 51 22 N
2 66 13 61 C
3 80 46 76 E
4 79 69 83 E
Fit the model:
In [7]: mod = smf.ols(formula='Lottery ~ Literacy + Wealth + Region', data=df)
In [8]: res = mod.fit()
In [9]: print(res.summary())
OLS Regression Results
==============================================================================
Dep. Variable: Lottery R-squared: 0.338
Model: OLS Adj. R-squared: 0.287
Method: Least Squares F-statistic: 6.636
Date: Tue, 28 Feb 2017 Prob (F-statistic): 1.07e-05
Time: 21:36:08 Log-Likelihood: -375.30
No. Observations: 85 AIC: 764.6
Df Residuals: 78 BIC: 781.7
Df Model: 6
Covariance Type: nonrobust
===============================================================================
coef std err t P>|t| [0.025 0.975]
-------------------------------------------------------------------------------
Intercept 38.6517 9.456 4.087 0.000 19.826 57.478
Region[T.E] -15.4278 9.727 -1.586 0.117 -34.793 3.938
Region[T.N] -10.0170 9.260 -1.082 0.283 -28.453 8.419
Region[T.S] -4.5483 7.279 -0.625 0.534 -19.039 9.943
Region[T.W] -10.0913 7.196 -1.402 0.165 -24.418 4.235
Literacy -0.1858 0.210 -0.886 0.378 -0.603 0.232
Wealth 0.4515 0.103 4.390 0.000 0.247 0.656
==============================================================================
Omnibus: 3.049 Durbin-Watson: 1.785
Prob(Omnibus): 0.218 Jarque-Bera (JB): 2.694
Skew: -0.340 Prob(JB): 0.260
Kurtosis: 2.454 Cond. No. 371.
==============================================================================
Warnings:
[1] Standard Errors assume that the covariance matrix of the errors is correctly specified.
Categorical variables¶
Looking at the summary printed above, notice that patsy determined
that elements of Region were text strings, so it treated Region as a
categorical variable. patsy‘s default is also to include an
intercept, so we automatically dropped one of the Region categories.
If Region had been an integer variable that we wanted to treat
explicitly as categorical, we could have done so by using the C()
operator:
In [10]: res = smf.ols(formula='Lottery ~ Literacy + Wealth + C(Region)', data=df).fit()
In [11]: print(res.params)
Intercept 38.651655
C(Region)[T.E] -15.427785
C(Region)[T.N] -10.016961
C(Region)[T.S] -4.548257
C(Region)[T.W] -10.091276
Literacy -0.185819
Wealth 0.451475
dtype: float64
Examples more advanced features patsy‘s categorical variables
function can be found here: Patsy: Contrast Coding Systems for
categorical variables
Operators¶
We have already seen that “~” separates the left-hand side of the model from the right-hand side, and that “+” adds new columns to the design matrix.
Removing variables¶
The “-” sign can be used to remove columns/variables. For instance, we can remove the intercept from a model by:
In [12]: res = smf.ols(formula='Lottery ~ Literacy + Wealth + C(Region) -1 ', data=df).fit()
In [13]: print(res.params)
C(Region)[C] 38.651655
C(Region)[E] 23.223870
C(Region)[N] 28.634694
C(Region)[S] 34.103399
C(Region)[W] 28.560379
Literacy -0.185819
Wealth 0.451475
dtype: float64
Multiplicative interactions¶
”:” adds a new column to the design matrix with the product of the other two columns. “*” will also include the individual columns that were multiplied together:
In [14]: res1 = smf.ols(formula='Lottery ~ Literacy : Wealth - 1', data=df).fit()
In [15]: res2 = smf.ols(formula='Lottery ~ Literacy * Wealth - 1', data=df).fit()
In [16]: print(res1.params)
Literacy:Wealth 0.018176
dtype: float64
In [17]: print(res2.params)
Literacy 0.427386
Wealth 1.080987
Literacy:Wealth -0.013609
dtype: float64
Many other things are possible with operators. Please consult the patsy docs to learn more.
Functions¶
You can apply vectorized functions to the variables in your model:
In [18]: res = smf.ols(formula='Lottery ~ np.log(Literacy)', data=df).fit()
In [19]: print(res.params)
Intercept 115.609119
np.log(Literacy) -20.393959
dtype: float64
Define a custom function:
In [20]: def log_plus_1(x):
....: return np.log(x) + 1.
....:
In [21]: print(res.params)
Intercept 115.609119
np.log(Literacy) -20.393959
dtype: float64
Namespaces¶
Notice that all of the above examples use the calling namespace to look for the functions to apply. The namespace used can be controlled via the eval_env keyword. For example, you may want to give a custom namespace using the patsy.EvalEnvironment or you may want to use a “clean” namespace, which we provide by passing eval_func=-1. The default is to use the caller’s namespace. This can have (un)expected consequences, if, for example, someone has a variable names C in the user namespace or in their data structure passed to patsy, and C is used in the formula to handle a categorical variable. See the Patsy API Reference for more information.
Using formulas with models that do not (yet) support them¶
Even if a given statsmodels function does not support formulas, you
can still use patsy‘s formula language to produce design matrices.
Those matrices can then be fed to the fitting function as endog and
exog arguments.
To generate numpy arrays:
In [22]: import patsy
In [23]: f = 'Lottery ~ Literacy * Wealth'
In [24]: y, X = patsy.dmatrices(f, df, return_type='dataframe')
In [25]: print(y[:5])
Lottery
0 41.0
1 38.0
2 66.0
3 80.0
4 79.0
In [26]: print(X[:5])
Intercept Literacy Wealth Literacy:Wealth
0 1.0 37.0 73.0 2701.0
1 1.0 51.0 22.0 1122.0
2 1.0 13.0 61.0 793.0
3 1.0 46.0 76.0 3496.0
4 1.0 69.0 83.0 5727.0
To generate pandas data frames:
In [27]: f = 'Lottery ~ Literacy * Wealth'
In [28]: y, X = patsy.dmatrices(f, df, return_type='dataframe')
In [29]: print(y[:5])
Lottery
0 41.0
1 38.0
2 66.0
3 80.0
4 79.0
In [30]: print(X[:5])
Intercept Literacy Wealth Literacy:Wealth
0 1.0 37.0 73.0 2701.0
1 1.0 51.0 22.0 1122.0
2 1.0 13.0 61.0 793.0
3 1.0 46.0 76.0 3496.0
4 1.0 69.0 83.0 5727.0
In [31]: print(smf.OLS(y, X).fit().summary())
OLS Regression Results
==============================================================================
Dep. Variable: Lottery R-squared: 0.309
Model: OLS Adj. R-squared: 0.283
Method: Least Squares F-statistic: 12.06
Date: Tue, 28 Feb 2017 Prob (F-statistic): 1.32e-06
Time: 21:36:08 Log-Likelihood: -377.13
No. Observations: 85 AIC: 762.3
Df Residuals: 81 BIC: 772.0
Df Model: 3
Covariance Type: nonrobust
===================================================================================
coef std err t P>|t| [0.025 0.975]
-----------------------------------------------------------------------------------
Intercept 38.6348 15.825 2.441 0.017 7.149 70.121
Literacy -0.3522 0.334 -1.056 0.294 -1.016 0.312
Wealth 0.4364 0.283 1.544 0.126 -0.126 0.999
Literacy:Wealth -0.0005 0.006 -0.085 0.933 -0.013 0.012
==============================================================================
Omnibus: 4.447 Durbin-Watson: 1.953
Prob(Omnibus): 0.108 Jarque-Bera (JB): 3.228
Skew: -0.332 Prob(JB): 0.199
Kurtosis: 2.314 Cond. No. 1.40e+04
==============================================================================
Warnings:
[1] Standard Errors assume that the covariance matrix of the errors is correctly specified.
[2] The condition number is large, 1.4e+04. This might indicate that there are
strong multicollinearity or other numerical problems.