Behavioral Response to an Anti Malaria Spraying Campaign, with Evidence from Eritrea Pedro Carneiro (UCL)
Andrea Locatelli (UCL)
Tewolde Gebremeskel (NMCP Eritrea)
Joseph Keating (Tulane)
[email protected]
AEL, 25 June 2011
Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
1 / 20
Research question
In the fight against Malaria: Is it counterproductive to introduce spraying campaigns, in areas with widespread bed-net coverage? Does the introduction of spraying campaigns lead to “perverse” behavioral responses, which may hinder Malaria eradication?
Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
2 / 20
Motivation
Technologies to fight Malaria: ITNs (bed-nets), IRS (spraying) and environmental management (eg draining stagnant water). Lengeler (RBM, 2011): introducing IRS where ITNs are widely available may hinder their acceptability, reducing take-up and use. “Malaria fatigue” (RBM, 2008): policy-induced Malaria reduction may induce loss of interest in Malaria. Should Malaria make a comeback, severe epidemics may ensue. Entire population at risk; especially children under 5 and pregnant women.
Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
3 / 20
Spraying intervention
Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
4 / 20
Preview
Results: The introduction of spraying campaigns did not lead to any “perverse” behavioral response. It actually promoted take-up and use of bed-nets. Explanation: The spraying campaign itself conveyed information on the importance of Malaria prevention. Increased awareness of Malaria threat led to increased ownership and use of bed-nets.
Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
5 / 20
Literature
P. Nikolov (2011): Availability of ARV drugs seems to distort incentives for consistent HIV preventive behavior. P. Dupas (2011): Provision of information can affect health-seeking behavior (Malaria, water, HIV). A. de Paula et al. (2011): Changes in beliefs of being HIV+ induce changes in risky behavior. Better information on HIV status reduces the transmission rate. A. Tarozzi et al. (2010): Promoting micro-loans to purchase bed-nets did not induce any “perverse” behavioral response. J. Keating et al. (2011): Cannot conclude that spraying offers additional protection, compared to bed-nets alone.
Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
6 / 20
Theoretical framework
Agents have an anti-Malaria technology, entailing disutility. Behavioral response to the introduction of a new, free preventive health technology that causes no disutility to users. Bed-nets and spray are (imperfect) substitutes. Imperfect information on Malaria prevalence. Source of information: can observe the introduction of IRS. Government is committed to eradicate Malaria.
Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
7 / 20
Predictions
After the introduction of IRS: Perfect info: average net use will decrease or remain unchanged. Imperfect info: average net use may even increase, if agents update beliefs about malaria burden, when they observe IRS campaign. We have assumed that spray and bed-nets are (imperfect) substitutes. Spray and bed-nets may be perceived as (imperfect) complements. Average net use will increase or remain unchanged.
Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
8 / 20
Data
Eritrea, Gash Barka Zone. Control: ITNs and LHM. Treatment: ITNs and LHM + IRS. Malaria season: Jul–Dec. Peak: Sep–Oct. IRS intervention: Jun–Jul, 2009. Data collection: 6–15 Oct, 2009. 115 villages, with 1,617 hh and 7,895 obs. No baseline data.
Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
9 / 20
Randomization protocol
List of malarious villages. Villages randomly selected for study. Distance check: min 5km. Villages randomly allocated to treatment or control. Large villages were segmented; one segment randomly chosen. 15 hh chosen for interviews using tables of random numbers.
Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
10 / 20
Randomization checks Variables (Y) ALL 1- Female 2- Usually lives here 3- Stayed there last night 4- Age
N
Mean
St. Dev.
p-value
7,826 7,740 7,709 7,880
0.52 0.98 0.96 22.17
0.5 0.15 0.2 19.35
0.722 0.206 0.113 0.484
RESPONDENTS ONLY 7- Age 8- Ever attended school 9- Only primary school 10- Literate 11- Muslim religion 12- Tigre tribe 13- Married
1,616 1,615 296 1,615 1,610 1,615 1,609
41.74 0.19 0.76 0.19 0.81 0.48 0.93
15.13 0.39 0.43 0.39 0.39 0.5 0.25
0.492 0.832 0.481 0.639 0.377 0.05* 0.348
Note: p-value of test β = 0 in Yi = βTi + εi . Obs. clustered at village level. *** p<0.01, ** p<0.05, * p<0.1. Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
11 / 20
Randomization checks (2) Variables (Y) N BY HOUSEHOLD 14- Hh size 1,617 15- # members under 5 1,616 16- # members under 18 1,616 17- Source of drinking water: Public tap 1,615 Unprotected well 1,615 18- Has any toilet 1,615 19- Has radio 1,615 20- Firewood is main fuel 1,601 21- Has no window 1,556 23- # sleeping rooms 1,615 24- # sleeping spaces 1,615
Mean
St. Dev.
p-value
4.89 0.83 2.64
2.29 0.92 1.97
0.239 0.705 0.471
0.43 0.24 0.06 0.25 0.95 0.32 1.38 4.53
0.49 0.43 0.24 0.43 0.23 0.47 0.77 2.4
0.893 0.721 0.629 0.797 0.248 0.939 0.969 0.39
Note: p-value of test β = 0 in Yi = βTi + εi . Obs. clustered at village level. *** p<0.01, ** p<0.05, * p<0.1.
Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
12 / 20
Malaria information and awareness Variables 1. Mosquitoes mentioned among Malaria vectors 2. Malaria is a problem in community 3. Children mentioned among most affected by Malaria 4. Pregnant women mentioned among most affected 5. Heard/saw messages about ITNs in past 6 months 6. Heard/saw messages on early seeking behavior 7. Heard/saw messages on environmental management
(1) Treatment 0.908 (0.289) 0.726 (0.446) 0.863 (0.344) 0.367 (0.482) 0.484 (0.500) 0.537 (0.499) 0.450 (0.498)
(2) Control 0.854 (0.353) 0.670 (0.471) 0.788 (0.409) 0.365 (0.482) 0.469 (0.499) 0.501 (0.500) 0.387 (0.487)
(3) β 0.0305* (0.016) 0.035 (0.035) 0.068*** (0.019) -0.0143 (0.024) -0.00050 (0.038) 0.019 (0.040) 0.029 (0.036)
Notes: Columns 1 and 2 report means for treatment and control groups, with standard deviations in parentheses. β in column 3 represents the treatment effect, estimated using probit regressions, for which marginal effects are reported. Additional controls: Tigre tribe, Muslim and subzone dummies. *** p<0.01, ** p<0.05, * p<0.1. Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
13 / 20
Ownership and use of bed-nets Variables 1. Number of nets owned by household 2. Number of ITNs owned by household 5. How much is willing to pay for a net, having none 6. Reported net use (of each household member) 7. Number of observed nets used the night before 8. Number of observed nets left unused the night before 9. Number of owned nets left unused the night before
(1) Treatment 1.774 (1.279) 1.444 (1.206) 24.346 (22.390) 0.429 (0.495) 1.384 (1.214) 0.676 (0.993) 0.756 (1.038)
(2) Control 1.575 (1.207) 1.278 (1.126) 23.296 (23.823) 0.380 (0.486) 1.164 (1.054) 0.736 (1.001) 0.818 (1.057)
(3) β 0.214** (0.0996) 0.176* (0.0926) 1.564 (3.126) 0.0329 (0.0317) 0.186** (0.0877) 0.0152 (0.0626) 0.0118 (0.0689)
Notes: Columns 1 and 2 report means for treatment and control groups, with standard deviations in parentheses. β in column 3 represents the treatment effect, estimated using LS regressions. Treatment effect for net use computed using probit regression, for which marginal effects are reported. Additional controls: Tigre tribe, Muslim and subzone dummies. *** p<0.01, ** p<0.05, * p<0.1. Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
14 / 20
Other risk-mitigating behaviors Variables 1. Household keeps livestock >100m from home 2. Household covers stored water 3. Respondent does anything to prevent mosquito bites 4. Respondent mentions using net 5. Respondent mentions burning coils 6. Respondent mentions using spray 9. Respondent mentions draining stagnant water
(1) Treatment 0.807 (0.395) 0.942 (0.234) 0.834 (0.372) 0.680 (0.467) 0.225 (0.418) 0.025 (0.156) 0.106 (0.309)
(2) Control 0.776 (0.417) 0.953 (0.212) 0.804 (0.397) 0.649 (0.478) 0.211 (0.409) 0.021 (0.143) 0.120 (0.325)
(3) β 0.068** (0.031) -0.027 (0.018) -0.006 (0.025) 0.011 (0.029) 0.003 (0.022) 0.010 (0.008) -0.022 (0.018)
Notes: Columns 1 and 2 report means for treatment and control groups, with standard deviations in parentheses. β in column 3 represents the treatment effect, estimated using probit regressions, for which marginal effects are reported. Additional controls: Tigre tribe, Muslim and subzone dummies. *** p<0.01, ** p<0.05, * p<0.1. Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
15 / 20
Het. T.E. on knowledge that mosquitoes are vector
Subsample: Treatment T x ndvi=1
T x ndvi=2
Work T x work
N
Y=1(Mosquitoes are a Malaria vector) (2) (3) (4) All Working age Working age men 0.0302* 0.020 0.0468** 0.111* (0.016) (0.025) (0.023) (0.058) -0.035 (0.043) [0.6704] 0.0641*** (0.024) [0.0005]*** 0.034 0.041 (0.024) (0.040) -0.061 -0.154** (0.045) (0.077) [0.7791] [0.4006] 1,597 1,597 1,504 515 (1) All
(5) Working age women 0.036 (0.025)
0.039 (0.031) 0.001 (0.054) [0.4213] 937
Note: Probit regressions, reporting marginal effects. *** p<0.01, ** p<0.05, * p<0.1. Additional controls include: (Gender in 1–3), Tigre tribe dummy, Muslim dummy, subzone dummies. Observations clustered at village level. Robust standard errors in parentheses. P-value for the F-test interaction+treatment=0 in square brackets.
Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
16 / 20
Heterogeneous T.E. on Malaria awareness (1) All
Subsample: Treatment
0.035 (0.035)
T x ndvi=1
T x ndvi=2
Work T x work
N
1,567
Y=1(Malaria is a problem) (2) (3) (4) All Working age Working age men 0.052 -0.026 -0.018 (0.072) (0.043) (0.081) -0.027 (0.085) [0.5251] -0.028 (0.113) [0.7514] -0.034 -0.003 (0.045) (0.078) 0.126*** 0.100 (0.049) (0.095) [0.0178]** [0.1655] 1,567 1,479 549
(5) Working age women -0.037 (0.047)
-0.032 (0.060) 0.131** (0.062) [0.1385] 918
Note: Probit regressions, reporting marginal effects. *** p<0.01, ** p<0.05, * p<0.1. Additional controls include: (Gender in 1–3), Tigre tribe dummy, Muslim dummy, subzone dummies. Observations clustered at village level. Robust standard errors in parentheses. P-value for the F-test interaction+treatment=0 in square brackets.
Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
17 / 20
Intra-household nets allocation
Subsample: Treatment Observations
Y=1(Reported net use, the night before the survey) (1) (2) (3) (4) (5) Children Youth Adult male Adult female Adult male under 5 aged 5-20 workers workers unemployed 0.017 0.033 0.084** 0.070 0.058 (0.039) (0.038) (0.042) (0.057) (0.056) 1,343 3,385 972 417 432
(6) Adult female unemployed 0.014 (0.040) 1,182
Note: Marginal effects estimated after probit regressions of dummy for net use. Additional controls include: Tigre tribe dummy, Muslim dummy, subzone dummies. Samples restricted as shown above. Observations clustered at village level. Robust standard errors in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01
Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
18 / 20
Main findings
1
No perverse behavioral response to IRS campaign.
2
IRS actually promoted bed-net take-up and use.
How? 3% more people know that mosquitoes are a Malaria vector. 12.6% more workers are aware of Malaria risk. Net (ITN) ownership increased by 21.4% (17.6%). It increased especially in hh with a working head. 0.19 more nets were used the night before the survey. Net use increased by 8% among workers. Young children and women: not negatively affected.
Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
19 / 20
Conclusions
1
Introduction of preventive-health measures can convey information.
2
This effect can outweigh perverse responses.
3
General result, applicable when new health intervention in introduced.
4
Further evidence that information can affect health-seeking behavior.
Andrea Locatelli (UCL)
Behavioral Response to IRS
AEL, 25 June 2011
20 / 20