Analyzing NYC High School SAT Scores

The SAT, or Scholastic Aptitude Test, is an exam that U.S. high school students take before applying to college. Colleges take the test scores into account when deciding who to admit, so it's fairly important to perform well on it.

The test consists of three sections, each of which has 800 possible points. The combined score is out of 2,400 possible points (while this number has changed a few times, the data set for our project is based on 2,400 total points). Organizations often rank high schools by their average SAT scores. The scores are also considered a measure of overall school district quality.

In this project I am going to investigate the correlation between SAT scores in NYC high schools and various demographics such as race, gender and income.

All of the data sets used in this project have been taken from

Here are the details all of the data sets I'll be using:

Basic research into New York, SAT's and the data has led to the following information:

Only high school students take the SAT, so I'llfocus on high schools. New York City is made up of five boroughs, which are essentially distinct regions. New York City schools fall within several different school districts, each of which can contains dozens of schools. Our data sets include several different types of schools. We'll need to clean them so that we can focus on high schools only. Each school in New York City has a unique code called a DBN, or district borough number. Aggregating data by district will allow us to use the district mapping data to plot district-by-district differences. To start with I am going to read the various data sets into pandas dataframes.

Read in the data

In [1]:
import pandas as pd
import numpy
import re

data_files = [

data = {}

for f in data_files:
    d = pd.read_csv("schools/{0}".format(f))
    data[f.replace(".csv", "")] = d

Read in the surveys

In [2]:
all_survey = pd.read_csv("schools/survey_all.txt", delimiter="\t", encoding='windows-1252')
d75_survey = pd.read_csv("schools/survey_d75.txt", delimiter="\t", encoding='windows-1252')
survey = pd.concat([all_survey, d75_survey], axis=0)

survey["DBN"] = survey["dbn"]

survey_fields = [
survey = survey.loc[:,survey_fields]
data["survey"] = survey

Add DBN columns

In [3]:
data["hs_directory"]["DBN"] = data["hs_directory"]["dbn"]

def pad_csd(num):
    string_representation = str(num)
    if len(string_representation) > 1:
        return string_representation
        return "0" + string_representation
data["class_size"]["padded_csd"] = data["class_size"]["CSD"].apply(pad_csd)
data["class_size"]["DBN"] = data["class_size"]["padded_csd"] + data["class_size"]["SCHOOL CODE"]

Convert columns to numeric

In [4]:
cols = ['SAT Math Avg. Score', 'SAT Critical Reading Avg. Score', 'SAT Writing Avg. Score']
for c in cols:
    data["sat_results"][c] = pd.to_numeric(data["sat_results"][c], errors="coerce")

data['sat_results']['sat_score'] = data['sat_results'][cols[0]] + data['sat_results'][cols[1]] + data['sat_results'][cols[2]]

def find_lat(loc):
    coords = re.findall("\(.+, .+\)", loc)
    lat = coords[0].split(",")[0].replace("(", "")
    return lat

def find_lon(loc):
    coords = re.findall("\(.+, .+\)", loc)
    lon = coords[0].split(",")[1].replace(")", "").strip()
    return lon

data["hs_directory"]["lat"] = data["hs_directory"]["Location 1"].apply(find_lat)
data["hs_directory"]["lon"] = data["hs_directory"]["Location 1"].apply(find_lon)

data["hs_directory"]["lat"] = pd.to_numeric(data["hs_directory"]["lat"], errors="coerce")
data["hs_directory"]["lon"] = pd.to_numeric(data["hs_directory"]["lon"], errors="coerce")

Condense datasets

In [5]:
class_size = data["class_size"]
class_size = class_size[class_size["GRADE "] == "09-12"]
class_size = class_size[class_size["PROGRAM TYPE"] == "GEN ED"]

class_size = class_size.groupby("DBN").agg(numpy.mean)
data["class_size"] = class_size

data["demographics"] = data["demographics"][data["demographics"]["schoolyear"] == 20112012]

data["graduation"] = data["graduation"][data["graduation"]["Cohort"] == "2006"]
data["graduation"] = data["graduation"][data["graduation"]["Demographic"] == "Total Cohort"]

Convert AP scores to numeric

In [6]:
cols = ['AP Test Takers ', 'Total Exams Taken', 'Number of Exams with scores 3 4 or 5']

for col in cols:
    data["ap_2010"][col] = pd.to_numeric(data["ap_2010"][col], errors="coerce")

Combine the datasets

In [7]:
combined = data["sat_results"]

combined = combined.merge(data["ap_2010"], on="DBN", how="left")
combined = combined.merge(data["graduation"], on="DBN", how="left")

to_merge = ["class_size", "demographics", "survey", "hs_directory"]

for m in to_merge:
    combined = combined.merge(data[m], on="DBN", how="inner")

combined = combined.fillna(combined.mean())
combined = combined.fillna(0)

Add a school district column for mapping

In [8]:
def get_first_two_chars(dbn):
    return dbn[0:2]

combined["school_dist"] = combined["DBN"].apply(get_first_two_chars)

Find correlations

In [9]:
correlations = combined.corr()
correlations = correlations["sat_score"]
SAT Critical Reading Avg. Score         0.986820
SAT Math Avg. Score                     0.972643
SAT Writing Avg. Score                  0.987771
sat_score                               1.000000
AP Test Takers                          0.523140
Total Exams Taken                       0.514333
Number of Exams with scores 3 4 or 5    0.463245
Total Cohort                            0.325144
CSD                                     0.042948
NUMBER OF SECTIONS                      0.362673
AVERAGE CLASS SIZE                      0.381014
SIZE OF SMALLEST CLASS                  0.249949
SIZE OF LARGEST CLASS                   0.314434
schoolyear                                   NaN
fl_percent                                   NaN
frl_percent                            -0.722225
total_enrollment                        0.367857
ell_num                                -0.153778
ell_percent                            -0.398750
sped_num                                0.034933
sped_percent                           -0.448170
asian_num                               0.475445
asian_per                               0.570730
black_num                               0.027979
black_per                              -0.284139
hispanic_num                            0.025744
hispanic_per                           -0.396985
white_num                               0.449559
rr_p                                    0.047925
N_s                                     0.423463
N_t                                     0.291463
N_p                                     0.421530
saf_p_11                                0.122913
com_p_11                               -0.115073
eng_p_11                                0.020254
aca_p_11                                0.035155
saf_t_11                                0.313810
com_t_11                                0.082419
eng_t_11                                0.036906
aca_t_11                                0.132348
saf_s_11                                0.337639
com_s_11                                0.187370
eng_s_11                                0.213822
aca_s_11                                0.339435
saf_tot_11                              0.318753
com_tot_11                              0.077310
eng_tot_11                              0.100102
aca_tot_11                              0.190966
grade_span_max                               NaN
expgrade_span_max                            NaN
zip                                    -0.063977
total_students                          0.407827
number_programs                         0.117012
priority08                                   NaN
priority09                                   NaN
priority10                                   NaN
lat                                    -0.121029
lon                                    -0.132222
Name: sat_score, Length: 67, dtype: float64

Plotting survey correlations

In [10]:
# Remove DBN since it's a unique identifier, not a useful numerical value for correlation.
In [27]:
%matplotlib inline
combined.corr()["sat_score"][survey_fields].sort_values(ascending = False)
<matplotlib.axes._subplots.AxesSubplot at 0x7fe170c05a90>

There are high correlations between N_s, N_t, N_p and sat_score. Since these columns are correlated with total_enrollment, it makes sense that they would be high.

It is more interesting that rr_s, the student response rate, or the percentage of students that completed the survey, correlates with sat_score. This might make sense because students who are more likely to fill out surveys may be more likely to also be doing well academically.

How students and teachers percieved safety (saf_t_11 and saf_s_11) correlate with sat_score. This make sense, as it's hard to teach or learn in an unsafe environment.

The last interesting correlation is the aca_s_11, which indicates how the student perceives academic standards, correlates with sat_score, but this is not true for aca_t_11, how teachers perceive academic standards, or aca_p_11, how parents perceive academic standards.

Exploring safety

In [13]:
combined.plot.scatter("saf_s_11", "sat_score");

There appears to be a correlation between SAT scores and safety, although it isn't thatstrong. It looks like there are a few schools with extremely high SAT scores and high safety scores. There are a few schools with low safety scores and low SAT scores. No school with a safety score lower than 6.5 has an average SAT score higher than 1500 or so.

Borough safety

In [14]:
# Compute the average safety score for each borough.
boros = combined.groupby("boro").agg(numpy.mean)["saf_s_11"]
Bronx            6.606577
Brooklyn         6.370755
Manhattan        6.831370
Queens           6.721875
Staten Island    6.530000
Name: saf_s_11, dtype: float64

It looks like Manhattan and Queens tend to have higher safety scores, whereas Brooklyn has low safety scores.

Racial differences in SAT scores

In [15]:
# make bar plot
race_fields = ["white_per", "asian_per", "black_per", "hispanic_per"]

It looks like a higher percentage of white or asian students at a school correlates positively with sat score, whereas a higher percentage of black or hispanic students correlates negatively with sat score. This may be due to a lack of funding for schools in certain areas, which are more likely to have a higher percentage of black or hispanic students.

In [16]:
# Make a scatter plot on low sat score and high values for hispanic
combined.plot.scatter("hispanic_per", "sat_score");

This scatter plot shows a negative relation with sat score where sat score moving downward with increasing hispanic_per

In [17]:
# Research any schools with a hispanic_per greater than 95%.
print(combined[combined["hispanic_per"] > 95]["SCHOOL NAME"])
44                         MANHATTAN BRIDGES HIGH SCHOOL
253                            MULTICULTURAL HIGH SCHOOL
Name: SCHOOL NAME, dtype: object

The schools listed above appear to primarily be geared towards recent immigrants to the US. These schools have a lot of students who are learning English, which would explain the lower SAT scores.

In [18]:
# Research any schools with a hispanic_per less than 10% and an average SAT score greater than 1800
print(combined[(combined["hispanic_per"] < 10) & (combined["sat_score"] > 1800)]["SCHOOL NAME"])
37                                STUYVESANT HIGH SCHOOL
151                         BRONX HIGH SCHOOL OF SCIENCE
187                       BROOKLYN TECHNICAL HIGH SCHOOL
Name: SCHOOL NAME, dtype: object

Many of the schools above appear to be specialized science and technology schools that receive extra funding, and only admit students who pass an entrance exam. This doesn't explain the low hispanic_per, but it does explain why their students tend to do better on the SAT -- they are students from all over New York City who did well on a standardized test.

Gender differences in SAT scores

In [20]:
gender_fields = ["male_per", "female_per"]

n the plot above, we can see that a high percentage of females at a school positively correlates with SAT score, whereas a high percentage of males at a school negatively correlates with SAT score. Neither correlation is extremely strong.

In [21]:
# Investigate schools with high SAT scores and a high female_per, make scatter plot.
combined.plot.scatter("female_per", "sat_score");

Based on the scatterplot, there doesn't seem to be any real correlation between sat_score and female_per. However, there is a cluster of schools with a high percentage of females (60 to 80), and high SAT scores.

In [23]:
# Research any schools with a female_per greater than 60% and 
# an average SAT score greater than 1700
print(combined[(combined["female_per"] > 60) & (combined["sat_score"] > 1700)]["SCHOOL NAME"])
5                         BARD HIGH SCHOOL EARLY COLLEGE
26                         ELEANOR ROOSEVELT HIGH SCHOOL
60                                    BEACON HIGH SCHOOL
302                          TOWNSEND HARRIS HIGH SCHOOL
Name: SCHOOL NAME, dtype: object

These schools appears to be very selective liberal arts schools that have high academic standards.

AP Exam Scores vs SAT Scores

In [25]:
# calculate the percentage of students in each school that took an AP exam.

combined["ap_per"] = combined["AP Test Takers "] / combined["total_enrollment"]

combined.plot.scatter(x='ap_per', y='sat_score');

It looks like there is a relationship between the percentage of students in a school who take the AP exam, and their average SAT scores. It's not an extremely strong correlation, though.

summery of results

Based on the research, these are our findings:

  • Schools with bigger share of Hispanic and Black students tend o perform poorly in SAT.
  • High expection in academic by student has got a positive impact in SAT.
  • Schools with safer environment tend to produce student that correlate positively with the SAT.Most students in these schools performs really good in SAT.
  • Passing in AP test, projects a good perfomance in the SAT.
  • Those students where english isn't the native language tend to perform poorly in SAT.


In this project, we cleaned,combined,visualized and analyzed different datasets containing informations about SAT scores and demographics in NYC public high schools.We have dig deep into SAT scores trying to find out how it correlates with diffirent fields.And here are the findings;

  • Shools with high percentage of females tend to do better in SAT compared to those with higer percentage of men. This indicates that female do better in SAT compared to men
  • White and Asian students tend to do better in SAT compared to Hispanic and Black students.Poor perfomance of Hispanic students, is caused by recent immigartion in US.
  • Students who always engage positively with the survey also do well in SAT.
  • Those who didn't do well in AP tests tend to do poorly in SAT tests as well

We also confirm that, schools that do enjoy safety, ignited a good perfomaance in SAT.

In [ ]: