Merge pull request #158 from pedrovma/main

Updating spreg to 1.7

.github/workflows/unittests.yml

@@ -30,7 +30,6 @@
        matrix:
          os: [ubuntu-latest]
          environment-file:
-           - ci/39-latest.yaml
            - ci/310-min.yaml
            - ci/310-latest.yaml
            - ci/311-latest.yaml

ci/310-latest.yaml

@@ -4,7 +4,7 @@ channels:
 dependencies:
   - python=3.10
   # required
-  - libpysal
+  - libpysal>=4.12.1
   - numpy>=1.3
   - pandas>=1.0
   - scipy>=0.11

ci/310-min.yaml

@@ -4,7 +4,7 @@ channels:
 dependencies:
   - python=3.10
   # required
-  - libpysal
+  - libpysal>=4.12.1
   - numpy>=1.3
   - pandas>=1.0
   - scipy>=0.11

ci/311-latest.yaml

@@ -4,7 +4,7 @@ channels:
 dependencies:
   - python=3.11
   # required
-  - libpysal
+  - libpysal>=4.12.1
   - numpy>=1.3
   - pandas>=1.0
   - scipy>=0.11

ci/312-latest.yaml

@@ -4,7 +4,7 @@ channels:
 dependencies:
   - python=3.12
   # required
-  - libpysal
+  - libpysal>=4.12.1
   - numpy>=1.3
   - pandas>=1.0
   - scipy>=0.11

docs/_static/references.bib

@@ -8,13 +8,13 @@
 
 @article{Anselin2021,
 	Author = {Anselin, Luc and Amaral, Pedro},
-	Journal = {},
-	Number = {},
-	Pages = {},
+	Journal = {Journal of Geographical Systems},
+	Number = {26},
+	Pages = {209–234},
 	Title = {Endogenous Spatial Regimes},
-	Url = {https://www.researchgate.net/publication/353411566_Endogenous_Spatial_Regimes},
+	Url = {https://doi.org/10.1007/s10109-023-00411-2},
 	Volume = {},
-	Year = {2021}}
+	Year = {2024}}
 
 @article{McMillen1992,
 	Abstract = {ABSTRACT. Commonly-employed spatial autocorrelation models imply heteroskedastic errors, but heteroskedasticity causes probit to be inconsistent. This paper proposes and illustrates the use of two categories of estimators for probit models with spatial autocorrelation. One category is based on the EM algorithm, and requires repeated application of a maximum-likelihood estimator. The other category, which can be applied to models derived using the spatial expansion method, only requires weighted least squares.},

docs/api.rst

@@ -7,6 +7,15 @@ API reference
 
 .. _models_api:
 
+Classic Models
+--------------
+
+.. autosummary::
+   :toctree: generated/
+
+    spreg.OLS
+    spreg.TSLS
+
 Spatial Regression Models
 -------------------------
 
@@ -15,10 +24,10 @@ These are the standard spatial regression models supported by the `spreg` packag
 .. autosummary::
    :toctree: generated/
 
-    spreg.OLS
+    spreg.GM_Lag
     spreg.ML_Lag
+    spreg.GMM_Error
     spreg.ML_Error
-    spreg.GM_Lag
     spreg.GM_Error
     spreg.GM_Error_Het
     spreg.GM_Error_Hom
@@ -28,11 +37,11 @@ These are the standard spatial regression models supported by the `spreg` packag
     spreg.GM_Endog_Error
     spreg.GM_Endog_Error_Het
     spreg.GM_Endog_Error_Hom
-    spreg.TSLS
-    spreg.ThreeSLS
+    spreg.NSLX
 
 Discrete Choice Models
 ----------------------
+
 .. autosummary::
     :toctree: generated/
 
@@ -60,6 +69,8 @@ Regimes models are variants of spatial regression models which allow for structu
     spreg.GM_Endog_Error_Regimes
     spreg.GM_Endog_Error_Hom_Regimes
     spreg.GM_Endog_Error_Het_Regimes
+    spreg.OLS_Endog_Regimes
+    spreg.GM_Lag_Endog_Regimes
     spreg.Skater_reg
 
 Seemingly-Unrelated Regressions
@@ -107,6 +118,7 @@ Diagnostic tests are useful for identifying model fit, sufficiency, and specific
     spreg.akaike
     spreg.schwarz
     spreg.condition_index
+    spreg.dwh
     spreg.jarque_bera
     spreg.breusch_pagan
     spreg.white

spreg/__init__.py

@@ -1,3 +1,4 @@
+
 import contextlib
 from importlib.metadata import PackageNotFoundError, version
 
@@ -17,13 +18,15 @@
 from .ml_error_regimes import *
 from .ml_lag import *
 from .ml_lag_regimes import *
+from .nslx import *
 from .ols import *
 from .ols_regimes import *
 from .panel_fe import *
 from .panel_re import *
 from .probit import *
 from .regimes import *
 from .skater_reg import *
+#from .spsearch import *
 from .sp_panels import *
 from .sputils import *
 from .sur import *

spreg/diagnostics.py

@@ -1,8 +1,7 @@
 """
-Diagnostics for regression estimations.
-
+Diagnostics for regression estimations. 
+        
 """
-
 __author__ = (
     "Luc Anselin luc.anselin@asu.edu, Nicholas Malizia nicholas.malizia@asu.edu "
 )
@@ -39,7 +38,7 @@
 ]
 
 
-def f_stat(reg, df=0):
+def f_stat(reg,df=0):
     """
     Calculates the f-statistic and associated p-value for multiple
     coefficient constraints :cite:`Greene2003`.
@@ -50,7 +49,7 @@ def f_stat(reg, df=0):
     ----------
     reg             : regression object
                       output instance from a regression model
-    df              : number of coefficient constraints
+    df              : number of coefficient constraints 
                       (zero constraint for last df coefficients in betas)
 
     Returns
@@ -101,15 +100,15 @@ def f_stat(reg, df=0):
     utu = reg.utu  # (scalar) residual sum of squares
     # default case, all coefficients
     if df == 0:
-        r = k - 1
+        r = k-1
         predy = reg.predy  # (array) vector of predicted values (n x 1)
         mean_y = reg.mean_y  # (scalar) mean of dependent observations
         U = np.sum((predy - mean_y) ** 2)
-    else:  # F test on last df coefficients
+    else:     # F test on last df coefficients
         y = reg.y
         r = df
-        x0 = reg.x[:, :-r]
-        olsr = BaseOLS(y, x0)  # constrained regression
+        x0 = reg.x[:,:-r]
+        olsr = BaseOLS(y,x0)  # constrained regression
         rtr = olsr.utu
         U = rtr - utu
     fStat = (U / r) / (utu / (n - k))
@@ -1386,10 +1385,17 @@ def likratiotest(reg0, reg1):
         raise Exception("Missing or improper log-likelihoods in regression objects")
     if likr < 0.0:  # always enforces positive likelihood ratio
         likr = -likr
-    pvalue = stats.chisqprob(likr, 1)
-    likratio = {"likr": likr, "df": 1, "p-value": pvalue}
-    return likratio
 
+    # generalize to multiple parameters, e.g., spatial Durbin
+    df = reg1.k - reg0.k
+    if not(df > 0):
+        df = 1
+
+    #pvalue = stats.chisqprob(likr, 1)
+    pvalue = stats.chisqprob(likr, df)
+    #likratio = {"likr": likr, "df": 1, "p-value": pvalue}
+    likratio = {"likr": likr, "df": df, "p-value": pvalue}
+    return likratio
 
 def dwh(reg):
     """
@@ -1408,23 +1414,23 @@ def dwh(reg):
                 and associated p-value
 
     """
-    n = reg.n
-    ny = reg.yend.shape[1]  # number of endogenous variables
+    n = reg.n  
+    ny = reg.yend.shape[1]   # number of endogenous variables
     qq = reg.h  # all exogenous and instruments
     xx = reg.z  # all exogenous and endogenous
     # get predicted values for endogenous variables on all instruments
-    py = np.zeros((n, ny))
+    py = np.zeros((n,ny))
     for i in range(ny):
-        yy = reg.yend[:, i].reshape(n, 1)
-        ols1 = BaseOLS(y=yy, x=qq)
+        yy = reg.yend[:, i].reshape(n,1)
+        ols1 = BaseOLS(y=yy,x=qq)
         yp = ols1.predy
-        py[0:n, i] = yp.flatten()
+        py[0:n,i] = yp.flatten()
     nxq = sphstack(xx, py)
     # F-test in augmented regression
     ols2 = BaseOLS(y=reg.y, x=nxq)
     dwh = f_stat(ols2, df=ny)
-    return dwh
-
+    return dwh 
+    
 
 def _test():
     import doctest