For the joint model testing the effect of drought, only the seven upazilas for which we have reliable rainfall data were included in the analysis19. There were also too few events to categorize the drought by severity –in comparison to flood events where we were able to categorise the floods by their severity.
Nutrition indicators
The analysis (Table 15) shows that,
as far as the nutritional indicators are concerned -similar to the individual-event analysis – no clear overall picture emerges. The zlen indicator seems to suggests that drought has a negative effect on child growth -all the values were negative and two are statistically significant: five and nine months after the event (p=0.013 for both); but the same relationship does not exist for zwfl. For zwfl the sign switched several times between the different lagged periods (positive,
negative, and then positive again) with two significant values –one negative at the seven months and one positive at the nine month lag.
The coefficients of the maternal BMI and the dietary diversity indexes dd and DD indicator had inconsistent and insignificant values Dietary diversity for mothers and children were negative just after the events, then positive (and even statistically significant) after seven months. This suggests that both children and mother diets were relatively more diversified in communities affected by drought than in the control communities after seven months.
Food security and food price indicators
Table 16 summarizes the results of the joint models for the food security and food price indicators. These indicators are more consistent than the nutrition indicators. They show in particular that communities affected by droughts are systematically facing (statistically) higher food prices
(including oil PO, rice PR and generic food basket PFB) than communities non affected by drought, and that the effect lasts for at least 9 months after the event starts. As a consequence the share of the food expense (FE) and the food loan (FL) indicators are also positive and statistically higher for drought-affected communities than for non-affected communities just after the event (p=0.008) and in months three and nine for FL (p<0.001 for both) and in months three and five after the event for FE (p<0.001 for both).
3.1.3. Cyclones
Individual-events
Four cyclones had been identified for which the conditions for rigorous DiD tests were fulfilled. These are the May 19, 1998; Nov 22, 1998; Oct 28, 2000;
and Nov 12, 2002 cyclones (cf. Table 8 above). Table 17 summarizes the results of the DiD tests on nutrition indicators run individually on each of these events20.
Nutrition indicators
Like for the two other types of events analysed so far (flood and drought) the fact that maternal BMI and DD and child dd were not collected during all the rounds does limit the analysis to zlen and zwfl indicators21. None of these two indicators seems however to show any consistency in terms of statistical significant effects.
Food security and food price indicators
The analyses of the potential impact of individual cyclone events on food security and food price indicators are shown in Table 18. The picture that emerges is not necessarily clear.
Some indicators behave as expected.
For instance the FE indicator (food expenditure) is primarily positive for the cyclones of May 1999, Oct 2000 and Nov 2002, suggesting that households affected by cyclones spend a larger amount of their income on food. This however is not necessarily confirmed by the price indicators (PR, PO, and PFB):
20 There were two methods through which these comparisons were done due to data limitations. For the May 1998 cyclone, as only a few rounds of data collection took place before event onset, this period was selected as the before period compared to the times after the event without controlling for seasonal variation. This method assumes that differences existing in the early part of 1998 were consistent with that experienced in other seasons. During the later 1998, 2000, and 2002 cyclones, full – or nearly full years- year of data collection was available before the event, the same season in the year prior to event onset was used as the before period enabling us to control for seasonal variation. This method implies that the seasonal variation experienced the year before onset was typical. This is a stronger assumption so whenever possible the second type of analysis will be presented though both methods were undertaken and were found to be largely consistent.
21 The Oct 2000 cyclone does have data for maternal BMI and DD and child dd but the absence of these information for any of the 3 other cyclone events rules out any comparison, thus limiting greatly the reliability of conclusions that would be drawn from this particular event.
Table 17. Cyclones individual event models(a) – nutrition indicators
zlen zwfl Maternal BMI child dd Maternal DD
coef p-value coef p-value coef p-value coef p-value coef p-value event
+3m +5m +7m +9m event +3m +5m +7m +9m event +3m +5m +7m +9m event +3m +5m +7m +9m
-0.134 -0.076 -0.010 0.009 -0.138
0.055 0.035 0.007 0.153 0.155 -0.012 0.050 -0.034 0.094 -0.063 0.139 0.000 -0.010 -0.011
0.072 0.257 0.081 0.122 0.064
-0.021 -0.016 0.063 0.123 0.087 0.043 -0.118 -0.085 -0.089 0.042 -0.004 0.068 -0.054 0.056
-0.058 0.191 0.419 0.026 -0.053 0.162 -0.053 -0.084 -0.191
1.487 -0.835 -1.277 -0.441 0.390
0.002 0.042 0.004
0.138 -0.784 -1.248 0.055 -0.150
0.012 0.002
0.030
0.000
Legend: Only statistically significant p-values are indicated (in bold), other non-significant p-values are omitted. Coefficient values highlighted in light red indicate cases for which the DiD test suggests a worsening situation in relation to nutrition, that is, a lower zlen, zwfl, BMI, child dd or maternal DD in the affected (treatment) communities, compared to the non-affected (control) communities. Dark red values indicate cases where the worsening of the nutritional situation is statistically significant.
zlen: Length/Height for age z-score; zwfl: weight for Length/Height z-score; BMI: Maternal body mass index; dd: Child dietary diversity; DD: Maternal dietary diversity. The notation +3m; +5m, +7m, and +9m refer to 3-month, 5-month, 7-month, and 9-month lagged periods respectively, while ‘event’
refers to the period during or just after the event.
(a) The following variables were also included in the models (but not shown in the table) to control for individual and household effects: age and sex of child; mother age and education; livelihood strategies (farmer; labour; transport; salary; business); family size; birth order; source for drinking; use of latrine –see definitions in section 2.5.2.
May 1998Nov 1998Oct 2000Nov 2002
Not collected in 1998
Not collected in 1998 Round not completed
none show any pattern corroborating that prices were higher following the cyclones events. As far as the FL (food loans) indicator is concerned, some positive and significant differences are observed during the May 1998 event and after the Nov 1998 event but a series of significant negative differences were also observed during and after the Oct 2000 event.
Table 18. Cyclone individual-event models(a) – food security and price indicators
zlen zwfl Maternal BMI child dd Maternal DD
coef p-value coef p-value coef p-value coef p-value coef p-value event
Legend: Only statistically significant p-values are indicated (in bold), other non-significant p-values are omitted. Coefficient values highlighted in light red indicate cases for which the test suggests a worsening situation in relation to food security, that is, a higher FL, FE, PR, PO, or PFB in the affected (treatment) communities, compared to the non-affected (control) communities. Dark red values indicate cases where the worsening of the food security or food price situation is statistically significant.
FL: food loan; FE: food expenditure; PR: rice price; PO: soybean oil price; PBF: price of food basket. The notation +3m; +5m, +7m, and +9m refer to 3-month, 5-month, 7-month, and 9-month lagged periods respectively, while ‘event’ refers to the period during or just after the event.
(a) The following variables were also included in the models (but not shown in the table) to control for individual and household effects: age and sex of child; mother age and education; livelihood strategies (farmer; labour; transport; salary; business); family size; birth order; source for drinking; use of latrine –see definitions in section 2.5.2.
May 1998Nov 1998Oct 2000Nov 2002
Missing Round
Table 19. Cyclone event joint models(a) – nutrition indicators zlen all events
zwfl all
zlen 50%
zwfl 50%
zlen 75%
zwfl 75%
zlen wind 40+
zwfl wind 40+
coef