Academic Impacts of Hurricane Sandy

The impact of Hurricane Sandy on
students at the City University of New
York
1
SIMON MCDONNELL
COLIN CHELLMAN
GILJAE LEE
DAVID CROOK
OFFICE OF POLICY RESEARCH
THE CITY UNIVERSITY OF NEW YORK
Hurricane Sandy
2
 October 29th 2012
 Over 40 New
Yorkers died
 CUNY:




4 campuses damaged
10 campuses used for
displaced NYers
Classes disrupted for
a week
Hardship
withdrawals
Questions
3
 How many students were impacted?
 Who were they?
 Where were they?
 Did students who lived in the surge zone do worse as
a result of being exposed the storm?

Were key measures of academic progress worse for students
living in the surge zone relative to similar students?
Outline
4
 Our research questions
 Some related literature
 Data & Methodology
 PSM Groups:
DiD
 Additional regressions analysis

 Results
 Fall 2011 FTF -> Fall 2012 V peers
 Conclusions & next steps
Literature - Potential Disruptions to Academic
Progress
5
Examples
Type of Impact
Academic
Non-Academic
Type of Disruption
Academic
Stopping out associated with lower
likelihood to graduate with a BA
Non-Academic
Increased education funding following
tsunami in Indonesia (2004)
Increased financial strain from student
loans repayment after dropping out
Changes in voting patterns following
Hurricane Andrew (1992)
 Many factors in student success:
 High school prep., full-time, gender, race, family history
 Natural disasters are historic, disruptive events
 Can be seen as disruptions in students’ academic and nonacademic lives
Literature - Potential Disruptions to Academic
Progress
6
Examples
Type of Impact
Academic
Non-Academic
Type of Disruption
Academic
Stopping out associated with lower
likelihood to graduate with a BA
Non-Academic
Increased education funding following
tsunami in Indonesia (2004)
Increased financial strain from student
loans repayment after dropping out
Changes in voting patterns following
Hurricane Andrew (1992)
 Disasters: Hurricane Katrina (Herlihy and Phillips,
2009), Indian ocean tsunami (Pelupessy et al),
school shootings etc.
 Family disruptions (Ver Ploeg, 2002)
 Drugs/alcohol (Arria et al, 2013)
 Little on higher ed. outcomes
Data
7
 Student-level data from Office of Institutional
Research and Assessment at CUNY;
 The Primary Land Use Tax Lot Output (PLUTO)
geospatial dataset;
 Inundation flood maps developed by FEMA;
 American Community Survey and decennial census
data from the U.S. Census Bureau.

Bad identification of impacted students?

Too inclusive?
Initial Results
8
 Fall 2012 snapshot:
 17,000 of 225,000 NYC CUNY students (undergrads and
grads) in surge zone – 8% of all students
 Brooklyn (50%), Manhattan (20%), Queens (18%), Staten
Island (8%), and the Bronx (3%).
 These numbers were used to reach out to potentially
impacted students
Methodology 1
9
 PSM – similar to CUNY ASAP methodology
 Nearest neighbor, caliper (.20 SD), non-replacement
 One Cohort over time
 Fall 2011 FTF  Retained to Fall 2012
GPA Diff (Semester 3- Semester 1)
 Credits Withdrawn Diff (Semester 3 – Semester 1)


Big caveat:
Retained to fall 2012 – more resilient students
 Fall 2012 students – similar results…

Fall 2011 Students  Retained to Fall 2012
10
Associate Students
Baccalaureate Students
Non-Surge Zone
Surge Zone
Non-Surge Zone
Surge Zones
Total Students (fall
2011)
19,958
1,665
8,745
766
Total Students
(retained to fall 2012)
12,915
(64.7%)
1,015
(60.9%)
7,599
(86.9%)
658
(85.9%)
Methodology 2
11
 DiD as Average Treatment Effect on the Treated
 ACADi represents the difference in GPA score and credits
withdrawn between semesters 1 and 3 for each student i.

ATT = ∑ i (ACADNon-Surge*Semester3 - ACADNon-Surge*Semester1) ∑ i (ACADSurge*Semester3 - ACADSurge*Semester1 )
 Additional regression analysis on matched groups
 Same as surge with additional controls
 Boro, college and major fixed effects
Methodology 3
12
 Group membership (AA & BA Separately):
 SURGEit = β0 + β1SEDit + β2ECOLit + β3CITit + β5HOODit + β5BOROit + εit





SED: NYCHA, race, gender, educationally disadvantaged, Pell
status, dependency status
ECOL: No delay, fulltime, credits attempted (sem. 1),
SEEK/CD, remediation
CIT: citizenship status
HOOD: neighborhood age, gender, earnings, education, race,
building stock
BORO: Boro fixed effects
PSM Results
13
 Fall 2011 FTF  Retained to Fall 2012
 AA:
1,015 student in Surge zone (7.3%)
  799 matched to a nearest neighbor (12,915 Non-surge)


BA:
658 student in Surge zone (7.9%)
  536 matched to a nearest neighbor (7,599 N0n-surge)


Bias reduction for almost all IV’s & balanced samples
Selected Descriptives: Associate
14
Variable
In NYCHA BBL
No Pell Disbursed
Educationally Disadvantaged
No delay after high school
Participant in SEEK/CD Program
In Need of Any Remediation
Age as of FTF date
Asian female
Asian male
Black female
Black male
Hispanic female
Hispanic male
Unmatched/
Matched
U
M
U
M
U
M
U
M
U
M
U
M
U
M
U
M
U
M
U
M
U
M
U
M
U
M
Treated
(Surge)
Control
(Non-Surge)
Bias
Reduction
% Bias
0.203
0.173
0.294
0.305
0.737
0.737
0.808
0.809
0.056
0.173
0.254
0.289
0.771
0.762
0.769
0.796
44.7
0
8.9
3.7
-7.8
-5.8
9.5
3.1
0.035
0.036
0.637
0.633
19.656
19.568
0.047
0.049
0.065
0.065
0.194
0.204
0.130
0.140
0.141
0.139
0.090
0.093
0.030
0.036
0.665
0.647
19.920
19.835
0.082
0.044
0.088
0.071
0.161
0.194
0.122
0.126
0.218
0.133
0.161
0.100
3.3
0
-5.8
-2.9
-6.5
-6.5
-14
2
-9
-2.4
8.7
2.6
2.5
4.1
-20.1
1.6
-21.5
-2.3
100
58.9
25.5
67.7
100
49.8
-1.2
85.4
73.7
69.9
-64.4
91.8
89.4
Fall 2011 FTF  Fall 2012: Associate
15
Variable
Sample
Difference in GPA Sem. 3 – Sem. 1
Unmatched
-0.303
-0.171
-0.132 0.063 -2.09
ATT
-0.303
-0.163
-0.140
0.071
-1.96
Unmatched
2.173
2.260
-0.087
0.059
-1.48
ATT
2.171
2.289
-0.118
0.066
-1.80
Unmatched
2.476
2.431
0.045
0.053
0.85
ATT
2.474
2.452
0.021
0.060
0.35
Unmatched
1.057
0.898
0.159
0.158
1.01
ATT
1.052
0.922
0.130
0.178
0.73
Unmatched
1.768
1.558
0.211
0.150
1.41
ATT
1.766
1.560
0.206
0.167
1.23
0.712
0.660
0.052
0.090
0.58
0.714
0.638
0.076
0.098
0.78
GPA: Semester 3
GPA: Semester 1
Diff. in Credits W/D Sem. 3 – Sem. 1
Credits Withdrawn: Semester 3
Credits Withdrawn: Semester 1
Treated
Controls
Difference S.E.
T-stat
Unmatched
ATT
Fall 2011 FTF  Fall 2012: Baccalaureate
16
Variable
Sample
Difference in GPA Sem. 3 – Sem. 1
Unmatched
-0.152
-0.148
-0.004 0.053 -0.07
ATT
-0.159
-0.115
-0.043 0.063 -0.69
Unmatched
2.844
2.877
-0.033
0.058
-0.58
ATT
2.841
2.937
-0.096
0.068
-1.41
Unmatched
2.996
3.026
-0.030
0.045
-0.66
ATT
2.999
3.052
-0.053
0.053
-0.99
Unmatched
0.461
0.417
0.044
0.158
0.28
ATT
0.473
0.280
0.192
0.185
1.04
Unmatched
0.906
0.896
0.010
0.146
0.07
ATT
0.905
0.787
0.118
0.166
0.71
0.445
0.479
-0.034
0.090
-0.38
0.432
0.507
-0.074
0.107
-0.69
GPA: Semester 3
GPA: Semester 1
Diff. in Credits W/D Sem. 3-Sem. 1
Credits Withdrawn: Semester 3
Credits Withdrawn: Semester 1
Treated
Controls
Difference S.E.
T-stat
Unmatched
ATT
What does this mean?
17
GPA Values - Current
Grade
Undergraduate
A+
A
AB+
B
BC+
C
4.0
4.0
3.7
3.3
3.0
2.7
2.3
2.0
D
1.0
F
WU
0
0
FIN
0
FAB
0
Additional Regression Analysis
18
 Academic Outcome (AA & BA Separately):
 ACADit = β0 + β1SURGEit + β2SEDit + β3ECOLit + β3CITit + β5HOODit +
β5HSit + β5FXit + εit
 GPA Differences and Credit Withdrawn Differences

Changed Major between semester 1 and semester 3
 Testing impact of additional boro, college and major fixed
effects
Regression Results: Fall 2011 FTF  Fall 2012
(Associate)
19
Associate Students: GPA DIFFERENCE (Sem. 1 – Sem. 3)
Model 1:
BORO
FX
Model 3:
Model 4:
Model 2A:
Model 3A:
Model 4A:
-0.168**
-0.194***
-0.198***
-0.183**
-0.207***
-0.211***
SED
+
+
+
+
+
+
ECOL
+
+
+
+
+
+
CIT
+
+
+
+
+
+
N’HOOD
+
+
+
+
+
+
HS
+
+
+
+
+
+
+
+
Lived in Surge
Boro FX
-0.134**
Model 2:
+
+
College FX
+
+
Major FX
+
+
+
+
+
+
N
1,598
1,467
1,467
1,467
1,467
1,467
1,467
R2
0.005
0.074
0.080
0.081
0.087
0.093
0.094
Regression Results: Fall 2011 FTF  Fall 2012
(Associate)
20
Associate Students: CREDITS W/D DIFFERENCE
Model 1:
BORO
FX
Lived in Surge
0.155
Model 2:
Model 3:
0.243
Model 4:
0.305*
Model 2A:
0.316*
0.252
Model 3A:
0.341**
Model 4A:
0.334**
SED
+
+
+
+
+
+
ECOL
+
+
+
+
+
+
CIT
+
+
+
+
+
+
N’HOOD
+
+
+
+
+
+
HS
+
+
+
+
+
+
+
+
Boro FX
+
+
College FX
+
+
Major FX
N
R2
1,598
0.007
1,467
1,467
0.068
0.062
1,467
0.073
+
+
+
+
+
+
1,467
1,467
1,467
0.088
0.093
0.095
Conclusions
21
 Small, and in many cases, insignificant impact on
our students (so far)


But more resilient students are… more resilient.
Interventions by CUNY and other actors?


Our sample more resilient than average?


Suspension of classes? Outreach?
Perhaps our students are resilient – look at 9/11 for instance
Bad identification of impacted students?
Too inclusive?
 Work with administrators etc to get a more nuanced sample
 Look at longer time frame to confirm treatment and control
groups follow same path

Next Steps
22
 Should we include a cross sectional of the fall 2012
entering cohort?
 Where should we aim this paper at?