diff-in-diff function.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import statsmodels.api as sm\n",
    "from mpl_toolkits.mplot3d import Axes3D\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def did2(df_parent,lower, upper, final_grade,  initial_grade, dummy_col, \n",
    "         treat_total, treat_normalized_new, num_L, num_U):\n",
    "    \n",
    "    \"\"\"\n",
    "    This function carries out difference-in-differences as a function of number of ADU attendance and range of initial grades.\n",
    "    \n",
    "    1. It first requests what number of ADU attendance, num_attend, do you want to look at?  It then looks for that number in \n",
    "    the total ADU column and populates a new column (treat_normalized_new) with 1s where num_attend was found in Total ADU.\n",
    "    \n",
    "    2.  Then, based on this DF, it extracts,all the observations between \n",
    "    a range of grades, lower and upper.  This is df.\n",
    "    \n",
    "    3.  Now that it has a subset DF with new T/C new column,  it then extracts a new DF that includes Total ADu ==0 (control) and\n",
    "    T/C new ==1.  This then creates a DF with the same control group each time we change the num_attend.  (We don't want the control g\n",
    "    group changing every time we change the treatment number.)\n",
    "    \n",
    "    4.  Now that we have our final DF, we need to get it into a form for calculating the DID.  We take the updated DF and then\n",
    "    subset to include only PHYS1015A, Mid-Year, T/C for example.   Takes in df of [Final Grade, Initial Grade, Treat/Control].  It then\n",
    "    rearranges the DF so that it is in the form to carry out difference-in-differences.\n",
    "    \n",
    "    It outputs the DID coefficient plus the p-value.\n",
    "    \n",
    "    imports:\n",
    "    \n",
    "    import statsmodels.api as sm\n",
    "    import pandas as pd \n",
    "    \n",
    "    \n",
    "    dummy_col = what are we controlling for?  Grades, or attendance?  If grades, then this will be Mid-Year.\n",
    "    If attendance, this will be Perc Videos (for video watching) or Total Attendance (for ADU tuts)\n",
    "    \n",
    "    attend = number of ADU sessions attended\"\"\"\n",
    "    ## new section that modifies the T/C column\n",
    "    \n",
    "    ## STEP 1:  CHOOSE NUM_ATTEND AS THOSE THAT ARE IN THE TREATMENT GROUP\n",
    "    df_parent[treat_normalized_new] = 0   # create new column populated with zeros\n",
    "    for k in df_parent[treat_total].index:\n",
    "        if dfTest1['2nd Sem'][k] >= num_L and dfTest1['2nd Sem'][k] < num_U:   # this line allows us to choose a range of \n",
    "            ## adu attendance.  You could just do 1 and 2 say, if you just wanted 1.  \n",
    "            df_parent[treat_normalized_new][k]= 1\n",
    "            \n",
    "    ## end of section that modifies T/C column\n",
    "    \n",
    "    \n",
    "    ## STEP 2:  CHOOSE THE RANGE OF INITIAL GRADE\n",
    "    df = df_parent[(df_parent[dummy_col]>lower) & (df_parent[dummy_col] <=upper) ]  # if controlling for grades\n",
    "    #df = df_parent[(df_parent[dummy_col]>lower) & (df_parent[dummy_col] <=upper) | (df_parent[dummy_col] == 0)]  # if controlling\n",
    "    # for video watching.  There needs to be a range of video watching plus the control \n",
    "    #df = df_parent[(df_parent[dummy_col]== attend) | (df_parent[dummy_col]==0)]\n",
    "    \n",
    "    \n",
    "    ## STEP 3:  SUBSET THE DF TO INCLUDE THE SAME CONTROL GROUP AND THE NEW TREAT GROUP\n",
    "    df = df[(df[treat_total] == 0) | (df[treat_normalized_new] == 1)]  ## important code.  Exract final df that includes\n",
    "    ## the same control group every time\n",
    "    \n",
    "    ## STEP 4:  EXTRACT ONLY THE THREE COLUMNS WE NEED FOR DID\n",
    "    \n",
    "    df = df.loc[:,[final_grade, initial_grade, treat_normalized_new]].reset_index(drop = True)\n",
    "    \n",
    "    #1. read in file df and place initial grades below final grades.  \n",
    "    # i.e place all grades in single column\n",
    "    \n",
    "    ## STEP 5: REARRANGE THE INITIAL GRADE TO FIT UNDER THE FINAL GRADES SO THAT WE HAVE A SINGLE COLUMN OF GRADES\n",
    "    df_2 = df.iloc[:,0].append(df.iloc[:,1]).reset_index(drop= True)\n",
    "    # Need to convert to DataFrame\n",
    "    df_3 = pd.DataFrame(df_2)\n",
    "    \n",
    "    # creates new column PrePost populated with 1s (indicating final time).\n",
    "    \n",
    "    ## STEP 6:  CREATE NEW COLUMN CALLED PREPOST THAT IS 1 IF FINAL GRADE AND 0 IF INITIAL GRADE\n",
    "    df_3['PrePost'] = 1  \n",
    "    # populate the bottom half with 0s (indicating initial time)\n",
    "    for k in df.index:\n",
    "        df_3['PrePost'][k+len(df)] = 0\n",
    "        \n",
    "        \n",
    "    # create T/C column and then duplicate it below itself   \n",
    "    ## STEP 7:  CREATE T/C COLUMN THEN DUPLICATE IT BELOW ITSELF\n",
    "    df_3['T/C'] = df.iloc[:,2]\n",
    "    for kk in df.index:\n",
    "        df_3['T/C'][kk+len(df)] = df.iloc[:,2][kk]   \n",
    "    \n",
    "    # Create DID column = T/C*PrePost\n",
    "    \n",
    "    ## STEP 8:  CREATE THE DID COLUMN THAT IS THE INTERACTION VALUE\n",
    "    df_3['DID'] = df_3['T/C']*df_3['PrePost']\n",
    "    \n",
    "    # Rename all the columns\n",
    "    ## STEP 9:  RENAME ALL THE COLUMNS\n",
    "    df_3.columns = ['Grades','PrePost','T/C','DID']\n",
    "    \n",
    "    treat_num = df_3['T/C'].sum()/2\n",
    "    \n",
    "   ## STEP 10:  CARRY OUT REGRESSION USING STATSMODELS.API\n",
    "    X = df_3[['PrePost','T/C','DID']]\n",
    "    y= df_3['Grades']\n",
    "    X = sm.add_constant(X)\n",
    "    model = sm.OLS(y,X).fit()\n",
    "    model.predict(X)\n",
    "    print_model = model.summary()\n",
    "    \n",
    "    return print(print_model), print('Treat number:' ,treat_num), print('ADU attendance:',num_L, 'to:', num_U-1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}