5.1 Descriptive statistics
A table with summary statistics is illustrated below, illustrating the overall pattern of the data. For instance, from our table, we can observe a total of 35,273 observations, among whom 18, 342 individuals live in non-camps, and 16,931 live in camps. Additionally, the occupational background of the individuals in the sample provides a better understanding of the population being studied (see Appendix C).
Table 1
The table shows the distribution and range of our data sample
|
Observations
|
Mean
|
Std deviations
|
Min
|
Max
|
Camps
|
16,931
|
-
|
-
|
-
|
-
|
Non-camps
|
18,342
|
-
|
-
|
-
|
-
|
Total
|
35,273
|
|
|
|
|
Employment (yes/no)
|
30,451
|
0.256
|
0.44
|
0
|
1
|
Employment earnings
(monthly in BDT)
|
2,560
|
4907.82
|
21730.16
|
0
|
1000000
|
Hours worked (per week)
|
3,172
|
35.85
|
23.10
|
0
|
100
|
Job search (yes/no)
|
9,842
|
0.40
|
0.49
|
0
|
1
|
Age (years)
|
34,423
|
23.80
|
17.24
|
0
|
100
|
Female
|
34,411
|
0.52
|
0.50
|
0
|
1
|
Married (yes/no)
|
26,599
|
0.59
|
0.49
|
0
|
1
|
Education (0- no edu,
1-primary, 2-secondary 3- tertiary)
|
30,436
|
0.65
|
0.48
|
0
|
1
|
Note: The distribution of the data. The first column shows the number of observations, followed by the mean (column 2), standard deviation (column 3), and range afterward. The observations are at the individual-level unit. The dependent variable measures employment status, earnings, hours worked, and search efforts. Measured using the CBPS question, (i) Did you work for livelihood in the last 7 days (ii) how much do you typically get paid in a month (iii) hours of the main job in last 7 days (iv) are you currently searching for a job?
To ensure difference in differences (DID) analysis is an accurate methodology for comparing treatment and control groups, both groups require to be verified to be similar in terms of factors that might influence economic outcomes, such as gender, age, education, and other factors.
Table 2
The table shows the number of observations and mean differences between treatment and control groups
|
Treatment
Observations
(1)
|
Control
Observations
(2)
|
Treatment
Mean
(3)
|
Control
Mean
(4)
|
Difference
Mean
(5)
|
Standard error
(6)
|
Demographic characteristics
|
|
|
|
|
|
|
Age (years)
|
14,720
|
11,788
|
20.85
|
21.91
|
1.07
|
0.22
|
Female (gender)
|
14,715
|
11,785
|
0.52
|
0.51
|
-0.01**
|
0.01
|
Married (yes)
|
10,248
|
8,443
|
0.52
|
0.52
|
-0.004
|
0.01
|
Education (yes)
|
12,401
|
10,128
|
0.66
|
0.70
|
0.04
|
0.01
|
Employment (yes/no)
|
12,411
|
10,129
|
0.21
|
0.24
|
0.03
|
0.01
|
Employment earnings
|
1,310
|
1,025
|
3674.99
|
5267.33
|
1592.345
|
938.15
|
Hours worked
|
1,546
|
1,514
|
34.49
|
37.37
|
2.88
|
0.83
|
Job search (yes/no)
|
2,266
|
1,399
|
0.39
|
0.41
|
0.02
|
0.02
|
Note: The average characteristics of the treatment and control and the differences before flooding in Aug 2019 are shown in column 5, followed by the Standard Error (S.E) in column 6. The dependent variable measures employment status, earnings, hours worked, and search. Measured using the CBPS question, (i) Did you work for livelihood in the last 7 days (ii) how much do you typically get paid in a month (iii) hours of the main job in the last 7 days (iv) are you currently searching for a job?
From Table 2, it can be observed that other than age, there is no significant difference between treatment and control groups regarding economic outcomes. This indicates that our differences in outcome between the treatment and control groups are due to flooding rather than preexisting differences among the treatment and control groups. The table below shows OLS regression results showing the relationship of the covariates with economic measures for this study, employment status, earnings, hours, and job search efforts.
Table 3
The relationship between demographic characteristics on economic measures
|
(1)
|
(2)
|
(3)
|
(4)
|
(5)
|
(6)
|
(7)
|
(8)
|
VARIABLES
|
Employment Status
|
+controls
|
Employment Earnings
|
+controls
|
Employment
Hours
|
+controls
|
Employment search
|
+controls
|
Age
|
0.00181***
|
0.000420**
|
90.56***
|
41.38***
|
0.125***
|
0.0419
|
-0.00265***
|
-0.000646
|
|
(0.000191)
|
(0.000201)
|
(19.91)
|
(15.59)
|
(0.0354)
|
(0.0360)
|
(0.000380)
|
(0.000486)
|
Female
|
-0.352***
|
-0.347***
|
111.2
|
-792.1
|
-17.07***
|
-18.21***
|
-0.422***
|
-0.417***
|
|
(0.00508)
|
(0.00555)
|
(484.8)
|
(575.3)
|
(0.773)
|
(0.787)
|
(0.00921)
|
(0.0117)
|
Married
|
0.165***
|
0.174***
|
1,522**
|
2,204***
|
-2.950***
|
-2.256**
|
0.0705***
|
0.0632***
|
|
(0.00561)
|
(0.00584)
|
(603.7)
|
(674.2)
|
(0.980)
|
(0.967)
|
(0.0101)
|
(0.0121)
|
Education
|
0.0670***
|
-0.00745
|
4,724***
|
3,272***
|
3.793***
|
1.886**
|
-0.0912***
|
-0.0310**
|
|
(0.00565)
|
(0.00674)
|
(839.9)
|
(666.1)
|
(0.823)
|
(0.822)
|
(0.00953)
|
(0.0128)
|
Rohingya
|
|
-0.0537*
|
|
-3,429***
|
|
-7.508**
|
|
0.000161
|
|
|
(0.0295)
|
|
(835.6)
|
|
(3.134)
|
|
(0.0527)
|
Lives in camp
|
|
-0.143***
|
|
-2,446***
|
|
-1.302
|
|
0.175***
|
|
|
(0.0295)
|
|
(661.8)
|
|
(3.137)
|
|
(0.0528)
|
Constant
|
0.288***
|
0.449***
|
-1,893**
|
3,318***
|
36.39***
|
42.95***
|
0.724***
|
0.481***
|
|
(0.00862)
|
(0.0112)
|
(899.2)
|
(596.2)
|
(1.516)
|
(1.632)
|
(0.0162)
|
(0.0242)
|
Observations
|
26,534
|
21,479
|
2,554
|
2,552
|
3,165
|
3,164
|
9,744
|
5,830
|
R-squared
|
0.201
|
0.244
|
0.012
|
0.029
|
0.121
|
0.150
|
0.183
|
0.251
|
Note. This table shows the results of the OLS regression analysis (columns 1 to 8). The observations are at the individual-level unit. The dependent variable measures employment status (columns 1 & 2), employment earnings (columns 3 & 4), employment hours worked (columns 5 & 6), and employment search (columns 7 & 8). Measured using the CBPS question, (i) Did you work for livelihood in the last 7 days (ii) how much do you typically get paid in a month (iii) hours of the main job in last 7 days (iv) are you currently searching for a job? In parentheses are the robust standard errors. ∗Significant at 10% level; ∗∗significant at 5% level; ∗∗∗significant at 1% level.
The results show that if an individual is a (i) woman, (ii) Rohingya, or (iii) lives in refugee camps, they are less likely to find employment and lower earnings in the coastal area of Cox Bazar, Bangladesh. These results portray an idea about the general conditions of the coastal population. Several plausible reasons exist for this phenomenon. For instance, Rohingyas are 5% less likely to find employment, earn 3,429 BDT lower monthly, and on average work 7.5 hours less than locals in a month. In addition, living in camps rather than non-camps reduces employment status by 14% and earnings by 2,433 BDT per month. In comparison to the roads and transportation in Cox Bazar town and nearby areas, the camps suffer from poor infrastructure and resources, making the camp population more vulnerable and affected by natural disasters such as floods and landslides (IOM, 2020). Camps are temporary residential solutions, and long-term residency is questionable; therefore, the instability and mobility raise concerns about hiring refugees from camps. Similarly, this provisional residency is a barrier to credit and finance, harming businesses and recovery (Ozturk, Serin and Altinoz, 2019). In addition, the camp layouts indicate a self-contained economy within camps with schools, healthcare facilities, markets, shops, and other institutes, and are prone to disruptions by natural disasters causing further unemployment. Lastly, most refugees depend on the agricultural sector, which is highly susceptible to flooding (see Table 1 Appendix C).
5.2 Benchmark analysis
The difference in differences (DID) analysis is used to investigate the economic impact of flooding across Cox Bazar, Bangladesh. In the first regression, dummy variables are assigned, either 1 or 0, to represent the treatment and control groups, areas with flooding and areas without flooding (see Table 3). The identification of upazilas and camps that underwent flooding is made using Landsat-8 satellite data. The DID results in Table 3 indicate that flooding reduces the probability of finding employment but positively influences employment earnings, hours, and eagerness to find a job. For instance, flood-affected individuals were 3% less likely to secure employment than those who did not (see column 2 of Table 4). Though flooding is found to improve earnings for employed individuals by 3,462 BDT per month, they are found to work 24.59 hours greater than the non-affected individuals (see columns 4 and 6). The last column indicates that flood-affected individuals are 10% more likely to look for employment due to unemployment caused by flooding. Referring to the literature, flood severity is expected to affect the relationship between flooding and integration. A finer measure would rather be a comparison of flood severity across the treatment and control areas rather than a binary indication of whether it flooded or not. Therefore, a continuous measure of flood severity is used to run the analysis by calculating the Normalised Difference Water Index (NDWI) from Landsat 8 satellite data (see section 4.1).
The DID coefficients remain consistent with previous results (see Table 5). For instance, a 1% increase in flood water harms employment probability by 0.1%. This implies that individuals who experience flooding are more likely to face unemployment than people who do not experience flooding. The majority of the coastal population of Cox Bazar depends on the farming and fishing industry and is prone to be harmed by natural disasters (i.e., flooding). Furthermore, damage to infrastructure, such as roads and bridges, might harm the daily commute of individuals harming employment. The coefficient from columns 4, 6, and 8 of Table 4 indicates that flooding positively affects employment earnings, hours worked, and the pursuit of jobs. The greater pay of the flood-affected employed individuals might be driven by the flood-induced lower supply of workers or an increase in pay as disaster benefits due to the struggle faced by the flood-affected individuals. Though the increase in hours worked indicates greater economic activity, the positive effect is contingent on factors such as (i) temporary nature of the increase, (ii) job insecurity, and (iii) labor misuse. The greater hours at work might be temporary and, towards recovering and rebuilding from the flood damage. For example, most of the coastal population in Cox Bazar depends on the agriculture industry, such as farming and fishing. In farming, greater labor hours are required arising from the inundation of farmland due to floods towards adaptation and mitigation of inundated farmland by water extraction and soil rehabilitation. Whereas, in other sectors, including fishing, employers allocate longer hours of work for employees to compensate for production losses due to flooding, (Paul and Rasid, 1993). In post-disaster situations where employment has suffered, individuals agree to work greater hours due to job insecurity, uncertainty surrounding their jobs, and fear of losing jobs (Kuvalekar and Lipnowski, 2019). Akhter and Kusakabe (2014) found Rohingya refugees get paid only half of the hourly wage of locals in Cox Bazar. Lastly, unemployment caused by flooding, or the urgency to earn during the natural calamity might have driven individuals to look for employment (see column 8).
In summary, flooding harms the probability of finding employment for the coastal population, relating to economic theory; such as Solow’s (1956) growth model, flooding harms factors of production capital and labor. For instance, flooding damages crops, disrupts the supply chain, inflates prices, and reduces consumption and businesses, along with temporary and permanent displacement of labor, causing labor shortages, and harming businesses and growth. Therefore, making individuals seek employment (see columns 2 and 8). Nonetheless, flooding improves earnings and increases hours given at work for employed individuals (see column 4). Though they constitute a much smaller number, the improvement in earnings might be driven by the greater opportunities in terms of labor, or as longer hours are contributed by the flood-affected individuals (see column 6). The longer hours at work might be driven by extra workload or the job insecurity caused by unemployment, however, it may also signify a shift in economic activity. The influence on economic harm or improvement will depend upon whether additional work is temporary or sustainable.
Table 4
The economic impact of flooding across the coastal population using binary treatment and control groups
|
(1)
|
(2)
|
(3)
|
(4)
|
(5)
|
(6)
|
(7)
|
(8)
|
VARIABLES
|
Employment Status
|
+controls
|
Employment Earnings
|
+controls
|
Employment
Hours
|
+controls
|
Employment search
|
+controls
|
Flood dummy
|
-0.0276***
|
-0.0235***
|
-1,592
|
-1,444
|
-2.878***
|
-2.742***
|
-0.0207
|
-0.0136
|
|
(0.00557)
|
(0.00587)
|
(1,039)
|
(1,010)
|
(0.835)
|
(0.783)
|
(0.0167)
|
(0.0146)
|
Post flooding
|
0.151***
|
0.0760***
|
4,319***
|
3,185**
|
-12.13***
|
-13.08***
|
-0.0751***
|
0.00671
|
|
(0.00890)
|
(0.00832)
|
(1,237)
|
(1,410)
|
(2.420)
|
(2.613)
|
(0.0159)
|
(0.0143)
|
Flood * Post
|
-0.0310**
|
-0.0278**
|
3,371**
|
3,462**
|
21.12***
|
24.59***
|
0.127***
|
0.100***
|
|
(0.0121)
|
(0.0112)
|
(1,600)
|
(1,600)
|
(4.017)
|
(4.307)
|
(0.0208)
|
(0.0185)
|
Age
|
|
0.00169***
|
|
75.55***
|
|
0.113***
|
|
-0.00289***
|
|
|
(0.000189)
|
|
(18.26)
|
|
(0.0354)
|
|
(0.000385)
|
Female
|
|
-0.354***
|
|
-179.7
|
|
-17.33***
|
|
-0.429***
|
|
|
(0.00523)
|
|
(410.1)
|
|
(0.771)
|
|
(0.00923)
|
Married
|
|
0.153***
|
|
1,608***
|
|
-2.852***
|
|
0.0629***
|
|
|
(0.00555)
|
|
(620.9)
|
|
(0.969)
|
|
(0.0101)
|
Education
|
|
0.0630***
|
|
4,240***
|
|
3.567***
|
|
-0.0901***
|
|
|
(0.00579)
|
|
(877.8)
|
|
(0.818)
|
|
(0.00953)
|
Constant
|
0.236***
|
0.297***
|
5,267***
|
-776.6
|
37.37***
|
38.36***
|
0.412***
|
0.710***
|
|
(0.00422)
|
(0.00984)
|
(1,003)
|
(563.7)
|
(0.615)
|
(1.597)
|
(0.0132)
|
(0.0187)
|
Observations
|
30,451
|
26,534
|
2,560
|
2,554
|
3,172
|
3,165
|
9,822
|
9,744
|
R-squared
|
0.021
|
0.206
|
0.008
|
0.018
|
0.009
|
0.132
|
0.008
|
0.192
|
Note. This table shows the results of the difference-in-differences method using a dummy treatment (columns 1 to 8). The observations are at the individual-level unit. The dependent variable measures employment status (columns 1 & 2), employment earnings (columns 3 & 4), employment hours worked (columns 5 & 6), and employment search (columns 7 & 8). Measured using the CBPS question, (i) Did you work for livelihood in the last 7 days (ii) how much do you typically get paid in a month (iii) hours of the main job in last 7 days (iv) are you currently searching for a job? In parentheses are the robust standard errors. ∗Significant at 10% level; ∗∗significant at 5% level; ∗∗∗significant at 1% level.
Table 5
The economic impact of flooding across the coastal population using flood severity measure
|
(1)
|
(2)
|
(3)
|
(4)
|
(5)
|
(6)
|
(7)
|
(8)
|
VARIABLES
|
Employment Status
|
+controls
|
Employment Earnings
|
+controls
|
Employment
Hours
|
+controls
|
Employment search
|
+controls
|
Flood severity
|
0.487***
|
0.458***
|
9,695*
|
7,246
|
-1.197
|
3.521
|
-0.901***
|
-0.474***
|
|
(0.0386)
|
(0.0390)
|
(5,226)
|
(5,683)
|
(4.903)
|
(4.529)
|
(0.107)
|
(0.0998)
|
Post flooding
|
0.137***
|
0.0636***
|
4,903***
|
3,696***
|
-8.442***
|
-8.802***
|
-0.0309**
|
0.0489***
|
|
(0.00749)
|
(0.00712)
|
(943.7)
|
(1,161)
|
(2.320)
|
(2.461)
|
(0.0129)
|
(0.0117)
|
Flood * Post
|
-0.0116**
|
-0.000538**
|
16,895*
|
19,722**
|
84.57***
|
97.87***
|
0.472***
|
0.233*
|
|
(0.0800)
|
(0.0738)
|
(9,306)
|
(9,450)
|
(23.84)
|
(25.84)
|
(0.132)
|
(0.123)
|
Age
|
|
0.00158***
|
|
77.90***
|
|
0.117***
|
|
-0.00286***
|
|
|
(0.000188)
|
|
(21.04)
|
|
(0.0355)
|
|
(0.000385)
|
Female
|
|
-0.355***
|
|
-496.4
|
|
-17.33***
|
|
-0.427***
|
|
|
(0.00520)
|
|
(384.7)
|
|
(0.774)
|
|
(0.00927)
|
Married
|
|
0.153***
|
|
1,580***
|
|
-2.858***
|
|
0.0613***
|
|
|
(0.00553)
|
|
(581.0)
|
|
(0.973)
|
|
(0.0101)
|
Education
|
|
0.0562***
|
|
4,124***
|
|
3.611***
|
|
-0.0868***
|
|
|
(0.00579)
|
|
(978.6)
|
|
(0.821)
|
|
(0.00964)
|
Constant
|
0.192***
|
0.265***
|
3,830***
|
-1,932**
|
36.00***
|
36.59***
|
0.448***
|
0.724***
|
|
(0.00340)
|
(0.00910)
|
(708.8)
|
(768.1)
|
(0.537)
|
(1.537)
|
(0.0103)
|
(0.0169)
|
Observations
|
30,451
|
26,534
|
2,560
|
2,554
|
3,172
|
3,165
|
9,822
|
9,744
|
R-squared
|
0.026
|
0.211
|
0.008
|
0.018
|
0.004
|
0.127
|
0.009
|
0.189
|
Note. This table shows the results of the difference-in-differences method using continuous treatment (columns 1 to 8). The observations are at the individual-level unit. The dependent variable measures employment status (columns 1 & 2), employment earnings (columns 3 & 4), employment hours worked (columns 5 & 6), and employment search (columns 7 & 8). Measured using the CBPS question, (i) Did you work for livelihood in the last 7 days (ii) how much do you typically get paid in a month (iii) hours of the main job in the last 7 days (iv) are you currently searching for a job? In parentheses are the robust standard errors. ∗Significant at 10% level; ∗∗significant at 5% level; ∗∗∗significant at 1% level.
5.3 Cash and non-cash assistance
As discussed in section 3.0, the current literature has examined how post-disaster relief or assistance can influence economic outcomes. Natural disasters like flooding can further worsen communities, especially vulnerable communities, such as the Rohingya refugees living across dreadful camps. To illustrate, most refugees in these camps live in temporary housing made using bamboo, tarpaulin sheets, mud, clay, and other locally available materials, unequipped to withstand flooding. Flood devastation can further harm already devasted Rohingyas refugees who have fled torture, conflict and lost mostly all belongings. Therefore, providing cash and non-cash assistance, such as food, shelter, water, and other essential services such as healthcare and education, might influence refugees' economic integration. Therefore, a two-stage least square analysis (2SLS) is used to investigate whether cash and non-cash assistance moderate the relationship between flooding and economic outcomes. The dummy measures of cash and non-cash assistance do not capture covid-19 assistance, and the individuals who received both cash and non-cash are excluded for this analysis.
Table 6
The first-stage regression for cash assistance
VARIABLES
|
(1)
Cash
|
(2)
Non-cash
|
Flood dummy
|
-0.00288
|
0.159***
|
|
(0.00392)
|
(0.00682)
|
Post flooding
|
0.0188***
|
0.174***
|
|
(0.00586)
|
(0.0139)
|
Flood * Post
|
0.0588***
|
0.00312*
|
|
(0.00879)
|
(0.0191)
|
Age
|
-0.000418***
|
-0.00397***
|
|
(0.000138)
|
(0.000243)
|
Female
|
-0.00559
|
-0.0238***
|
|
(0.00359)
|
(0.00644)
|
Married
|
0.00604
|
-0.0252***
|
|
(0.00404)
|
(0.00726)
|
Education
|
-0.0217***
|
-0.218***
|
|
(0.00421)
|
(0.00736)
|
Constant
|
0.102***
|
0.555***
|
|
(0.00684)
|
(0.0118)
|
Observations
|
25,507
|
21,449
|
R-squared
|
0.010
|
0.084
|
Note. This table shows the results of the first-stage regression. The dependent variable measures whether the participant received any assistance from government or non-governmental bodies. In parentheses are the robust standard errors. ∗Significant at 10% level; ∗∗significant at 5% level; ∗∗∗significant at 1% level.
Table 7
The second-stage regression for post-disaster cash and non-cash assistance
|
(1)
|
(2)
|
(3)
|
(4)
|
(5)
|
(6)
|
(7)
|
(8)
|
VARIABLES
|
Cash
Employment
Status
|
Non-cash
Employment
Status
|
Cash
Employment Earnings
|
Non-cash
Employment
Earnings
|
Cash
Employment
Hours
|
Non-cash
Employment
Hours
|
Cash
Employment
Search
|
Non-cash
Employment
Search
|
Cash/Non-cash dummy
|
0.605***
|
-0.0773***
|
56,493***
|
635.8
|
-104.7***
|
-8.295**
|
0.995***
|
0.301***
|
|
(0.0992)
|
(0.0225)
|
(16,471)
|
(2,847)
|
(34.59)
|
(4.011)
|
(0.144)
|
(0.0363)
|
Age
|
0.00205***
|
0.00150***
|
107.6***
|
93.00***
|
0.0811**
|
0.0883**
|
-0.00244***
|
-0.00163***
|
|
(0.000192)
|
(0.000208)
|
(17.86)
|
(26.33)
|
(0.0389)
|
(0.0395)
|
(0.000380)
|
(0.000397)
|
Female
|
-0.349***
|
-0.353***
|
225.6
|
133.8
|
-17.64***
|
-17.36***
|
-0.423***
|
-0.420***
|
|
(0.00528)
|
(0.00526)
|
(499.5)
|
(452.7)
|
(0.807)
|
(0.790)
|
(0.00919)
|
(0.00921)
|
Married
|
0.156***
|
0.165***
|
1,214*
|
1,534**
|
-2.316**
|
-3.076***
|
0.0579***
|
0.0694***
|
|
(0.00565)
|
(0.00544)
|
(667.0)
|
(629.3)
|
(1.003)
|
(0.984)
|
(0.0103)
|
(0.0101)
|
Education
|
0.0803***
|
0.0493***
|
5,671***
|
4,858***
|
1.492
|
1.950
|
-0.0684***
|
-0.0231*
|
|
(0.00620)
|
(0.00779)
|
(650.0)
|
(1,315)
|
(1.133)
|
(1.215)
|
(0.0100)
|
(0.0125)
|
Constant
|
0.223***
|
0.341***
|
-7,420***
|
-2,304
|
47.14***
|
41.82***
|
0.607***
|
0.510***
|
|
(0.0139)
|
(0.0176)
|
(1,157)
|
(2,475)
|
(3.936)
|
(3.030)
|
(0.0233)
|
(0.0303)
|
Observations
|
26,527
|
26,527
|
2,554
|
2,554
|
3,165
|
3,165
|
9,739
|
9,739
|
R-squared
|
0.202
|
0.202
|
0.014
|
0.012
|
0.123
|
0.122
|
0.187
|
0.189
|
Note. This table shows the results of the second-stage regression (columns 1 to 8). The observations are at the individual-level unit. The dependent variable measures employment status (columns 1 and 2), employment earnings (columns 3 and 4), employment hours worked (columns 5 and 6), and employment search efforts (columns 7 and 8). Measured using the CBPS question, (i) Did you work for livelihood in the last 7 days (ii) how much do you typically get paid in a month (iii) hours of the main job in the last 7 days (iv) are you currently searching for a job? In parentheses are the robust standard errors. ∗Significant at 10% level; ∗∗significant at 5% level; ∗∗∗significant at 1% level.
The first-stage regression results show individuals in flooded areas get greater cash and non-cash assistance. The results from the second stage regression indicate that cash assistance improves employment among flood-affected individuals; whereas, non-cash assistance harms employment. Individuals who suffered from flooding and received cash assistance were 61% more likely to find employment, whereas flood-affected recipients of non-cash assistance were 7% less likely to find employment, compared to individuals who did not receive any assistance (see columns 1 and 2 of Table 7). In addition, cash assistance is found to positively influence earnings. These are important revelations against the debate to restrict cash assistance and focus on non-cash assistance. Financial assistance helps as a buffer for individuals who have lost their homes and possessions due to flooding, providing for basic needs and expenses and helping financial stability. As discussed in section 3.1, according to the income effect, an increase in income might initiate more income-generating activities, such as investing more time at work, entrepreneurial activities, or even training and developing skills. This can be associated with findings from Boca and Sorrenti’s (2020) paper, where they found that recipients of cash assistance in low-income households had better labor force participation as they utilized the cash assistance towards activities that promote labor force participation. Similarly, cash assistance has been associated with better financial health and food access in poor households in Columbia (Londoño-Vélez and Querubin, 2022). Cash assistance has been related to positive economic impact and welfare consequences (Angelucci and De Giorgi, 2009; Cunha, De Giorgi, and Jayachandran, 2018; Filmer et al., 2021), and can be related to behavioral responses towards earnings, consumption, food security, child growth and schooling and women empowerment, and improvement in financial well-being (Baird, McIntosh and Ozler, 2011; Bastagli et al. 2016; Haushofer and Shapiro, 2016).Whereas non-cash assistance and employment, being inversely related, can be linked to the leisure effect of the substitution effect. For instance, when basic needs such as food, clothing, shelter, and others are fulfilled by government and non-governmental assistance, non-cash assistance alters the relative price of leisure. Individuals can avoid work and instead have more leisure time, described as the ‘welfare trap’ (Murray, 1982).
Both cash and non-cash assistance reduce hours at work (see columns 5 and 6 of Table 7). The financial cushion from cash assistance allows refugees to allocate more time to searching for a job or improvement in terms of hours or pay. For instance, Hagen-Zanker et al. (2017) studied the effect of cash assistance on Syrian refugees in Jordan and found cash assistance covers basic needs and provides more time for job search and labor market outcomes. Cash and non-cash assistance can issue a sense of security and stability post-disaster, making them more confident and optimistic, and fostering better work. This might indicate the decrease in hours worked. However, the fall in hours at work, and increase in job search efforts might also illustrate lower economic activity due to the leisure effect as the substitution effect suggests. However, the question used to capture the hours worked is asked to a much smaller sample who are currently employed.
Nonetheless, the improvement in employment status and earnings for the coastal population exemplify the economic benefits for cash assistance recipients. The 2SLS results exhibit cash assistance as a better form of assistance in terms of economic results in society through greater employment and earnings. However, to further verify the findings, the effectiveness of cash and non-cash assistance are tested across several well-being measures using propensity score matching. Propensity score matching would allow to observe outcomes of food access and health in the absence of receiving cash or non-cash assistance. For instance, individuals with similar characteristics are matched, and individuals who have received cash assistance are compared to those who have not received any. A similar analysis is done for non-cash assistance recipients. The parameter of interest is the average treatment effect of cash and non-cash assistance on the treated (ATT)
ATT = E[Y(1)|D = 1] − E[Y(0)|D = 1]
In this equation, E[Y(1)|D = 1] is the outcome value when the individual receives cash or non-cash assistance, and E[Y(0)|D = 1] is the outcome value when individuals do not receive cash or non-cash assistance. The Nearest neighbor propensity score matching technique is used to match individuals across demographic characteristics. To estimate the probability, a logit regression is run including a set of covariates previous model.
Table 8
Propensity score matching
|
N Treatment
(1)
|
N Control
(2)
|
Treatment
Mean
(3)
|
Control
Mean
(4)
|
ATT
(5)
|
Std error
(6)
|
t value
(7)
|
Cash
|
|
|
|
|
|
|
|
Food access
|
2256
|
21561
|
0.17
|
0.07
|
0.11
|
0.03
|
3.89
|
Good health
|
789
|
9536
|
0.73
|
0.67
|
0.06
|
0.05
|
1.21
|
Non-cash assistance
|
|
|
|
|
|
|
|
Food access
|
7127
|
13166
|
0.12
|
0.21
|
-0.09
|
0.03
|
-2.44
|
Good health
|
3979
|
6419
|
0.73
|
0.69
|
0.04
|
0.04
|
0.93
|
Note: The table shows the results of the propensity score matching analysis. Food access and good health are measured using the CBPS question, (i) Did you have smaller or fewer meals (ii) In general, how would you say your health is? The t-value shows the significance of the differences between the treatment and control groups.
The t-values from the table above indicate that those who receive cash assistance have better access to food and better health than those who do not receive cash assistance. However, individuals who receive non-cash assistance, though have better health, had no significant differences in terms of having access to food. The non-cash assistance in the form of food is perishable, and assigned to individuals who either have previously received it or are currently experiencing a shortage. In addition, selling non-cash assistance such as food to local markets is a customary event among the poor coastal population. Therefore, the apprehension of losing out on non-cash assistance, and the possibility of earning extra prompts the revelation of no improvement in access to food. The results indicate that cash assistance outperforms non-cash assistance, and individuals who receive cash assistance possess better well-being.
5.4 Healthcare facilities and integration
The assumption behind flooding having a lower negative impact on areas with healthcare facilities is developed from the notion that individuals in areas with healthcare facilities have better health than areas without close proximity to healthcare facilities. Therefore, to test the assumption correlation, and t-test is conducted between flooded and non-flooded areas.
Table 9
The correlation and mean difference between good health and health facilities
|
Correlation with
Health facility
(1)
|
Facility (yes)
Observations
(2)
|
Facility (no)
Observations
(3)
|
Facility (yes)
Mean
(4)
|
Facility (no)
Mean
(5)
|
Difference
Mean
(6)
|
Standard error
(7)
|
All
|
|
|
|
|
|
|
|
Good health
|
0.025*
|
7,813
|
2,610
|
0.74
|
0.72
|
-0.03**
|
0.01
|
Only camps
|
|
|
|
|
|
|
|
Good health
|
0.023*
|
2,451
|
2,610
|
0.74
|
0.72
|
-0.02**
|
0.01
|
Note: The table shows the results of the correlation between good health and healthcare facilities. The good health variable is measured using the CBPS question, (i) In general, how would you say your health is? The t-value shows the significance of the differences between the treatment and control groups.
Column (1) of the table above indicates a significant positive correlation between individuals in areas with healthcare facilities and health. The t-test results also show significant differences between areas with and without healthcare facilities (see column 6). Therefore, the assumption is justified, and to investigate the disparities of the impact of flooding across areas with and without healthcare facilities, a triple difference estimator is used (see Table 10).
Table 10
The impact of flooding across areas with and without healthcare facilities
|
(1)
|
(2)
|
(3)
|
(4)
|
VARIABLES
|
Employment
Status
|
Employment
Earnings
|
Employment
Hours
|
Employment
Search
|
Flood areas
|
0.0460***
|
732.2
|
0.387
|
-0.116***
|
|
(0.0102)
|
(471.6)
|
(1.768)
|
(0.0218)
|
Health care facility
|
0.174***
|
4,588***
|
8.992***
|
-0.144***
|
|
(0.00970)
|
(1,141)
|
(1.536)
|
(0.0217)
|
Post flooding
|
0.118***
|
4,419***
|
-4.649
|
0.0216
|
|
(0.0144)
|
(1,322)
|
(4.478)
|
(0.0224)
|
Flood * Post
|
-0.142***
|
-1,185
|
18.81***
|
0.197***
|
|
(0.0153)
|
(1,570)
|
(5.651)
|
(0.0227)
|
Flood * Health
|
-0.0736***
|
-2,468*
|
-2.857
|
0.146***
|
|
(0.0117)
|
(1,293)
|
(1.962)
|
(0.0231)
|
Health * Post
|
-0.0513***
|
-1,811
|
-9.731*
|
-0.0131
|
|
(0.0150)
|
(1,483)
|
(5.082)
|
(0.0230)
|
Flood * Health * Post
|
1.058***
|
38,111***
|
32.93
|
-1.169***
|
|
(0.101)
|
(10,236)
|
(39.04)
|
(0.128)
|
Age
|
0.00114***
|
68.08***
|
0.0863**
|
-0.00228***
|
|
(0.000187)
|
(17.51)
|
(0.0354)
|
(0.000380)
|
Female
|
-0.359***
|
-392.3
|
-17.67***
|
-0.409***
|
|
(0.00514)
|
(409.8)
|
(0.776)
|
(0.00936)
|
Married
|
0.152***
|
1,740***
|
-2.685***
|
0.0639***
|
|
(0.00549)
|
(638.8)
|
(0.962)
|
(0.00996)
|
Education
|
0.0322***
|
3,744***
|
2.691***
|
-0.0529***
|
|
(0.00587)
|
(810.5)
|
(0.816)
|
(0.00974)
|
Constant
|
0.194***
|
-3,908***
|
31.84***
|
0.763***
|
|
(0.0116)
|
(1,128)
|
(1.970)
|
(0.0231)
|
Observations
|
26,534
|
2,554
|
3,165
|
9,744
|
R-squared
|
0.227
|
0.023
|
0.145
|
0.213
|
Note. This table shows the results of the triple differences estimator (columns 1 to 4). The observations are at the individual-level unit. The dependent variable measures employment status (column 1), employment earnings (column 2), employment hours worked (column 3), and employment search efforts (column 4). Measured using the CBPS question, (i) Did you work for livelihood in the last 7 days (ii) how much do you typically get paid in a month (iii) hours of the main job in the last 7 days, and (iv) are you currently searching for a job? In parentheses are the robust standard errors. ∗Significant at 10% level; ∗∗significant at 5% level; ∗∗∗significant at 1% level.
The triple estimator (DDD) coefficient indicates that individuals in areas with flood-affected healthcare facilities have better employment and earnings than those in areas without healthcare facilities. For instance, individuals who live in a flood-affected area with a healthcare facility are 1.05% more likely to find employment and have 38,111 more earnings per month than flood-affected individuals not living in an area with healthcare facilities. Though there is no improvement in hours worked, individuals were less likely to look for employment as unemployment is lower. Individuals close to areas with healthcare facilities have easier access to healthcare and a healthier individual is understood to be more productive at work (Kirsten, 2010; Bubonya; Cobb-Clark and Wooden, 2017; Mottaghi, 2018). In addition, as healthcare facilities tend to maintain high standards of hygiene and cleanliness to avoid the spread of diseases, areas the proximity of such facilities have a higher degree of general cleanliness. Though there is no movement restriction to enjoy healthcare facilities in another upazila or camp, the already limited resources and overpopulated camps cause the access and quality of healthcare to diminish. According to IOM reports, refugees tend to avoid healthcare, and experience longer waiting times, limited services, and inadequate resources when placed further away from healthcare facilities (IOM, 2020). The results emphasize the presence of healthcare facilities in these areas for better integration.
5.5 Robustness checks
5.5.1 Falsification
A falsification test is conducted to validate the benchmark model by ensuring the conclusions are not based on flawed assumptions, i.e., certain conditions are changed to falsify the hypothesis being tested. For instance, to investigate the relationship between flooding and employment measures, false treatment groups were set by assigning camps and non-camps (camps 01E, 29, 30,32, and Pekua) with minimal rainfall and flooding as treatment and all other areas as control.
Table 11
Falsification test using false treatment and control groups
|
(1)
|
(2)
|
(3)
|
(4)
|
VARIABLES
|
Employment
Status
|
Employment
Earnings
|
Employment
Hours
|
Employment
Search
|
Flood severity
|
-0.0755***
|
-1,809
|
-1.575
|
0.0883***
|
|
(0.0102)
|
(1,439)
|
(1.505)
|
(0.0238)
|
Post flooding
|
0.0597***
|
4,943***
|
-2.117
|
0.0736***
|
|
(0.00580)
|
(1,637)
|
(2.145)
|
(0.0101)
|
Flood * Post
|
0.0189
|
583.8
|
17.41*
|
-0.109
|
|
(0.0199)
|
(4,662)
|
(9.229)
|
(0.0316)
|
Age
|
0.00173***
|
77.95*
|
0.121***
|
-0.00299***
|
|
(0.000190)
|
(41.02)
|
(0.0354)
|
(0.000380)
|
Female
|
-0.354***
|
-347.0
|
-17.10***
|
-0.431***
|
|
(0.00506)
|
(1,213)
|
(0.873)
|
(0.00925)
|
Married
|
0.154***
|
1,621
|
-2.927***
|
0.0612***
|
|
(0.00569)
|
(1,093)
|
(1.002)
|
(0.0101)
|
Education
|
0.0641***
|
4,236***
|
3.736***
|
-0.0946***
|
|
(0.00564)
|
(913.3)
|
(0.828)
|
(0.00963)
|
Constant
|
0.289***
|
-1,446
|
36.72***
|
0.699***
|
|
(0.00869)
|
(1,529)
|
(1.492)
|
(0.0163)
|
Observations
|
26,534
|
2,554
|
3,165
|
9,744
|
R-squared
|
0.207
|
0.017
|
0.122
|
0.188
|
Note. This table shows the results of the falsification test using the difference in differences estimation (columns 1 to 4). The observations are at the individual-level unit. The dependent variable measures employment status (column 1), employment earnings (column 2), employment hours worked (column 3), and employment search efforts (column 4). Measured using the CBPS question, (i) Did you work for livelihood in the last 7 days (ii) how much do you typically get paid in a month (iii) hours of main job in last 7 days and (iv) are you currently searching for a job? In parentheses are the robust standard errors. ∗Significant at 10% level; ∗∗significant at 5% level; ∗∗∗significant at 1% level.
The table above shows insignificant coefficients for the DID variable for all the dependent variables. As the false treatment group faced no flooding, as expected, no significant effect was found. Similarly, for triple difference estimation analysis, a falsification test was conducted. The camps with the highest number of households with no access to healthcare facilities are set as treatment; camps 10, 11,19, and 20 as camps with healthcare facilities, and the rest as areas without healthcare facilities.
Table 12
falsification test using false health care facility sub-sample.
|
(1)
|
(2)
|
(3)
|
(4)
|
VARIABLES
|
Employment
Status
|
Employment
Earnings
|
Employment
Hours
|
Employment
Search
|
Flood areas
|
-0.0255***
|
-1,513
|
-2.929***
|
-0.0139
|
|
(0.00596)
|
(1,042)
|
(0.791)
|
(0.0150)
|
Health care facility
|
-0.106***
|
-3,656***
|
-6.848**
|
0.0274
|
|
(0.0223)
|
(990.1)
|
(3.154)
|
(0.0463)
|
Post flooding
|
0.0792***
|
3,035**
|
-13.34***
|
0.00424
|
|
(0.00848)
|
(1,451)
|
(2.660)
|
(0.0147)
|
Flood * Post
|
-0.0277**
|
3,630**
|
25.15***
|
0.0961***
|
|
(0.0115)
|
(1,645)
|
(4.428)
|
(0.0190)
|
Flood * Health
|
0.0246
|
776.0
|
7.336
|
0.0898
|
|
(0.0347)
|
(1,050)
|
(5.830)
|
(0.0775)
|
Health * Post
|
-0.0787**
|
5,116
|
11.03***
|
0.0596
|
|
(0.0388)
|
(1,972)
|
(4.082)
|
(0.0624)
|
Flood * Health * Post
|
0.0644
|
-4,750**
|
-19.11*
|
-0.0274
|
|
(0.0539)
|
(2,416)
|
(9.887)
|
(0.0948)
|
Age
|
0.00162***
|
75.29***
|
0.111***
|
-0.00282***
|
|
(0.000189)
|
(18.28)
|
(0.0353)
|
(0.000385)
|
Female
|
-0.354***
|
-212.5
|
-17.37***
|
-0.429***
|
|
(0.00522)
|
(413.8)
|
(0.773)
|
(0.00924)
|
Married
|
0.154***
|
1,643***
|
-2.840***
|
0.0628***
|
|
(0.00554)
|
(626.3)
|
(0.969)
|
(0.0101)
|
Education
|
0.0602***
|
4,205***
|
3.468***
|
-0.0872***
|
|
(0.00581)
|
(867.6)
|
(0.818)
|
(0.00956)
|
Constant
|
0.303***
|
-633.4
|
38.67***
|
0.704***
|
|
(0.00992)
|
(556.2)
|
(1.605)
|
(0.0190)
|
Observations
|
26,534
|
2,554
|
3,165
|
9,744
|
R-squared
|
0.208
|
0.018
|
0.133
|
0.194
|
Note. This table shows the results of the falsification test, using the triple differences estimator (columns 1 to 4). The observations are at the individual-level unit. The dependent variable measures employment status (column 1), employment earnings (column 2), employment hours worked (column 3), and employment search efforts (column 4). Measured using the CBPS question, (i) Did you work for livelihood in the last 7 days (ii) how much do you typically get paid in a month (iii) hours of the main job in the last 7 days, and (iv) are you currently searching for a job? In parentheses are the robust standard errors. ∗Significant at 10% level; ∗∗significant at 5% level; ∗∗∗significant at 1% level.
The coefficient for the triple differences estimator (Flood * Health* Post) shows no significant effect on employment status and employment earnings. These results indicate that the study is not sensitive to the effects of the tested conditions, and the model used is reliable and can be trusted accurately to reflect the true relationship between flooding and economic and social integration.
5.5.2 Placebo
To further test whether the model is effective in determining the economic impact of flooding, a placebo test is run. Individuals are randomly allocated from the whole population into two groups, treatment, and control. The placebo analysis would not have represented a thorough comparison of treatment and control groups, and therefore insignificant coefficients were expected.
Table 12
Placebo test using randomly allocated treatment and control groups
|
(1)
|
(2)
|
(3)
|
(4)
|
VARIABLES
|
Employment
Status
|
Employment
Earnings
|
Employment
Hours
|
Employment
Search
|
Flood severity
|
0.00765
|
358.2
|
0.0822
|
0.0172
|
|
(0.00594)
|
(896.3)
|
(0.785)
|
(0.0146)
|
Post flooding
|
0.0642***
|
4,113**
|
5.075
|
0.0735***
|
|
(0.00779)
|
(2,093)
|
(3.116)
|
(0.0133)
|
Flood * Post
|
-0.00340
|
1,884
|
-11.25***
|
-0.0220
|
|
(0.0109)
|
(3,030)
|
(4.193)
|
(0.0185)
|
Age
|
0.00175***
|
80.01*
|
0.126***
|
-0.00298***
|
|
(0.000190)
|
(41.00)
|
(0.0353)
|
(0.000380)
|
Female
|
-0.353***
|
-236.8
|
-17.10***
|
-0.432***
|
|
(0.00507)
|
(1,211)
|
(0.871)
|
(0.00925)
|
Married
|
0.153***
|
1,579
|
-2.973***
|
0.0617***
|
|
(0.00569)
|
(1,092)
|
(1.001)
|
(0.0101)
|
Education
|
0.0658***
|
4,241***
|
3.787***
|
-0.0948***
|
|
(0.00564)
|
(913.6)
|
(0.827)
|
(0.00963)
|
Constant
|
0.276***
|
-1,875
|
36.39***
|
0.700***
|
|
(0.00913)
|
(1,587)
|
(1.532)
|
(0.0177)
|
Observations
|
26,534
|
2,554
|
3,165
|
9,744
|
R-squared
|
0.205
|
0.017
|
0.123
|
0.187
|
Note. This table shows the results of the placebo test using the difference in differences estimation (columns 1 to 4). The observations are at the individual-level unit. The dependent variable measures employment status (column 1), employment earnings (column 2), employment hours worked (column 3), and employment search efforts (column 4). Measured using the CBPS question, (i) Did you work for livelihood in the last 7 days (ii) how much do you typically get paid in a month (iii) hours of main job in the last 7 days and (iv) are you currently searching for a job?. In parentheses are the robust standard errors. ∗Significant at 10% level; ∗∗significant at 5% level; ∗∗∗significant at 1% level.
The results show no significant coefficients, justifying that the model represents the true economic effect of flooding. However, it is acknowledgeable that the DID estimates are prone to selection biases harming the external validity of the model. Referring to the data, normally representative CBPS data were used, where participants are chosen through random sampling based on several demographic characteristics to develop our experimental treatment and control groups based on their exposure to flooding. Due to the data not being randomized by participants’ previous exposure to flooding, there is the probability that the DID sample systematically excludes a subset of flood-affected individuals who are not orthogonal to our dependent variables, which might harm the external validity of our model. To address this issue, four subsets of full sample T1 and C1, T2 and C1, T1 and C2, and T2 and C2 are constructed by randomly drawing half individuals from treatment and control groups to test whether results still hold for these sub-samples. The external validity test results are illustrated in Appendix D. The results show significant DID interaction coefficients when the model is run for subsets T1 and C1 or T2 and C2. Adversely, when among both treatments (T1 and T2) or control (C1 and C2), insignificant coefficients are found. Lastly, due to the collection of the tracking data concurrently occurring during covid-19 pandemic, and having a lower number of participants from the same households in baseline data, additional control for individuals who have lost employment due to covid and weighs for sample size adjustment pre- and post-flooding to benchmark analysis for further robustness checks are added (See Appendix E). The DID coefficients remain significant further proving the robustness of the model.