Post on 18-Jan-2016
Information delivery, Nutrition and HIV Treatment: Evidence from a randomized field experiment on women living with HIV in UgandaPatrick Lubega, Frances Nakakawa, Gaia Narciso and Carol Newman
Information delivery, Nutrition and HIV Treatment: Evidence from a randomized field experiment on women living with HIV in Uganda
• The aim is to test the impact of interventions aimed at improving the nutrition of women living with HIV in Uganda.
• We are interested in testing:
1. Their impact on health and welfare outcomes
2. Their impact on behavior related to nutrition
3. Their impact on behavior related to self-employment activities
4. The mechanisms through which changes in behavior occur
• To achieve this we use a randomized controlled trial design
Background• Uganda’s national response to HIV/AIDS is recognised as strong and effective
• Prevalence rate is estimated at 7.4% (UNAIDS, 2014)
• Increased enrolment of people on ART from 570,373 in 2013 to 750,896 in 2014.
• However, an area in need of research is the relationship between HIV treatment outcomes and nutrition
• Ugandan Ministry for Health has produced a training manual to guide practitioners in the management of nutritional aspects of HIV care and provides information on when and how to use RUTFs in the setting of HIV and severe malnutrition.
• But, a significant percentage of those attending for treatment have mild to moderate malnutrition and do not receive supplementary feeding under the present criteria
Experimental designWe randomly selected 4 sub-regions of Uganda;
We randomly selected 24 clinics, 6 in each of the four sub-regions;
At baseline each clinic was visited for two days during the HIV; clinic to recruit participants;
On average 135 women recruited in each clinic;
Baseline survey instrument gathers information on:
» Personal characteristics of woman » Personal characteristics of family members» Food frequency questionnaire» Income » Agricultural production» Enterprise activity» Employment» Housing» Access to credit and savings behavior
Experimental design
• Two separate treatments are considered and randomization of baseline participants occurs at the clinic level;
• Each group includes 8 clinics and is evenly distributed across each of the 4 sub-regions.
Intervention
Group A Nutritional Information campaign
Group B Nutritional Information campaign + recipe demonstration
Group C Control group
Timeline
April-Sept 2014Baseline
Oct-Dec 2014Intervention 1
Jan – Mar 2015Intervention 2
Apr – May 2015Intervention 3/Evaluation 1
July-Aug 2015Evaluation 2
Details on interventions
Nutritional Information Campaign
‒ Delivered by clinic staff at each visit
‒ Content of campaign:
» Information leaflets and posters
» Hands off approach
» Allow clinics to do as they please in attempt to mimic typical campaigns by Ministry of Health
Treatment 1: Nutritional Information Campaign
Treatment 2: Recipe Demonstration
Locally sourced Home-made Nutritious Foods• Nutritionists working in each region to establish locally sourced ingredients
that can be used to prepare a Home-Made Nutritious Food.
• Products tested at Makerere University for nutrient content, taste and aesthetic appeal – similar to bringing product to market
• Intervention takes the form of cookery demonstrations
Outcomes
• Health and education
• Income and livelihoods
• Empowerment
• Anthropometrics
Using our RCT design we can obtain the average impact of a each intervention on outcomes for women in treated clinics by comparing them to the women in the control clinics
‒ Women in cookery clinics more likely to acknowledge that they received access to information.
‒ Positive impact of information only on number of meals per day only.
‒ Cookery has positive impact on number of snacks per day. This is consistent with trying the recipe.
‒ Cookery clinics more likely than information to try the recipe
Behavioural change due to information delivery
Health and education
‒ Lower probability of reporting illness in both information and cookery clinics – effect is bigger in cookery
‒ Lower proportion of children in the household reported as being sick in cookery (no effect of information alone)
‒ Cookery reduces the proportion of children that are absent from school because of failure to pay school fees (no effect of information alone)
Questions: What is mechanism? Is the better well-being of women in the cookery clinics impacting on incomes?
Income and livelihoods
• Cookery has positive impact on
‒ all types of personal income
‒ household income
‒ the decision to start an enterprise.
• This suggests that the increase in well-being associated with the cookery campaign leads to greater earning ability of women.
• This allows them to pay school fees and reduces the proportion of days that children are absent from school.
• The information campaign also increases the health of women possibly because they eat more frequent meals. There are no knock on effects for children or effects on other outcomes such as income.
Income and livelihoods
These results raise new questions
The fact that the same welfare outcomes are not observed in information clinics means that there is something different about the cookery clinics.
There are two possibilities:
i. they become super-nourished because they are eating the recipes
ii. there is something about the way the information is delivered that empowers women to work harder.
Empowerment
‒ We find an increase in control over personal resources in cookery clinics coupled with a decline in the joint control of personal finances
Anthropometrics
Decline in anthropometrics in cookery clinics.
Question: They are working harder so they are losing weight. But overall they are feeling better.
• Is it diversity in diet?
• Consumption of micronutrients?
• More analysis needed.
Conclusions
• Lower probability of reporting illness in both information and cookery clinics – effect is bigger in cookery
• Lower proportion of children in the household reported as being sick in cookery
• Cookery reduces the proportion of children that are absent from school because of failure to pay school fees (no effect of information alone)
• Cookery has positive impact on
‒ all types of personal income
‒ household income
‒ the decision to start an enterprise.
Conclusions
• This suggests that the increase in well-being associated with the cookery campaign leads to greater earning ability of women.
• This allows them to pay school fees and reduces the proportion of days that children are absent from school.
• The information campaign also increases well-being of women but with no knock on effects for children or other welfare outcomes
• Some evidence of increase in control over personal resources in cookery clinics coupled with decline in joint control of personal finances
• Decline in anthropometrics in cookery clinics
Thank You!
Appendix
Econometric specification
is the particular outcome variable of interest for woman i in time period t;
is a dummy indicator for whether woman i is in an information clinic;
is a dummy indicator for whether woman i is in a cookery clinic;
is a dummy indicator for the time period – zero for the baseline and one for the evaluations;
is a vector of time varying control variables;
are woman fixed effects.
1 1 2( * ) ( * )it t i t i t i itO Time Info Time Cook Time itφX
itO
tTime
iInfo
iCook
itφX
i
Behavioural change in relation to information contained in campaign
(1) (2) (3) (4) (5) (6) Access to
InformationMeals Snacks Litres of
waterTreated Water Recipe
Eval*info 0.071 0.182* 0.445 0.271 0.081 (0.176) (0.076) (0.116) (0.104) (0.338) Eval*cook 0.234*** 0.132 0.609* -0.052 0.038 0.534*** (0.000) (0.214) (0.012) (0.681) (0.083) (0.000)
Evaluation dummy Yes Yes Yes Yes Yes NoIndividual FE Yes Yes Yes Yes Yes NoControls Yes Yes Yes Yes Yes Yes Mean control baseline 0.824 2.464 1.125 2.202 0.793 N.A.
Observations 3,900 3,899 3,901 3,905 3,877 1,346R-squared 0.067 0.040 0.086 0.020 0.049 0.262Number of Ind. 1,966 1,966 1,966 1,966 1,966
Standard errors computed using wild bootstrapping to account for clustering at the clinic level (Cameron, Gelbrach, and Miller, 2008). p-values in parenthesis. *** p<0.01, ** p<0.05, * p<0.1
Health and education (1) (2) (3) (4) (5)
Sick % child sick % child absent school
Child absent - school fees
Child absent - sick
Eval*info -0.085* -0.058 -0.039 -0.047 0.015
(0.092) (0.116) (0.451) (0.262) (0.539)
Eval*cook -0.145** -0.072** -0.087 -0.048* 0.034
(0.012) (0.034) (0.136) (0.094) (0.186)
Evaluation dummy Yes Yes Yes Yes YesIndividual FE Yes Yes Yes Yes Yes
Controls Yes Yes Yes Yes Yes
Mean control baseline 0.272 0.102 0.362 0.227 0.172
Observations 3,897 2,800 2,800 2,800 2,800
R-squared 0.026 0.021 0.023 0.042 0.010
Number of Ind. 1,966 1,615 1,615 1,615 1,615
Standard errors computed using wild boostrapping to account for clustering at the clinic level (Cameron, Gelbrach, and Miller, 2008). p-values in parenthesis. *** p<0.01, ** p<0.05, * p<0.1
Income and livelihoods (1) (2) (3) (4) (5) (6) (7) Personal
incomeOther household
incomeWage income Crop income Livestock
incomeEnterprise
incomeOperates enterprise
Eval*info -11,742 -3,760 -2,958 9,182 -2,779 4,560 -0.027
(0.342) (0.631) (0.288) (0.326) (0.322) (0.511) (0.575)
Eval*cook 47,917*** 26,932* 6,621** 17,321* 2,545* 15,722** 0.283***
(0.000) (0.086) (0.022) (0.060) (0.052) (0.044) (0.002)
Evaluation dummy Yes Yes Yes Yes Yes Yes Yes
Individual FE Yes Yes Yes Yes Yes Yes Yes
Controls Yes Yes Yes Yes Yes Yes Yes
Mean control baseline 83,071 50,601 22,154 26,510 5402 19,157 0.262
Observations 3,910 3,910 3,910 3,910 3,910 3,910 3,892
R-squared 0.024 0.029 0.019 0.019 0.013 0.012 0.088
Number of Ind. 1,966 1,966 1,966 1,966 1,966 1,966 1,965
Standard errors computed using wild boostrapping to account for clustering at the clinic level (Cameron, Gelbrach, and Miller, 2008).p-values in parenthesis. *** p<0.01, ** p<0.05, * p<0.1
Empowerment (1) (2) (3) (4) (5) (6)
Makes decisions about personal income alone
Husband makes decisions about personal income
Jointly make decisions about personal income
Makes decisions about household
income alone
Husband makes decisions about
household income
Jointly make decisions about
household income
Eval*info 0.059 0.004 -0.050 0.106 0.012 -0.002
(0.190) (0.809) (0.266) (0.394) (0.929) (0.947)
Eval*cook 0.214*** -0.009 -0.189*** -0.160 0.067 0.034
(0.000) (0.400) (0.002) (0.136) (0.483) (0.727)
Evaluation dummy Yes Yes Yes Yes Yes Yes
Individual FE Yes Yes Yes Yes Yes Yes
Controls Yes Yes Yes Yes Yes Yes
Mean control baseline 0.794 0.050 0.156 0.213 0.328 0.431
Observations 1,197 1,197 1,197 961 961 961
R-squared 0.159 0.028 0.152 0.103 0.026 0.026
Number of Ind. 611 611 611 587 587 587
Standard errors computed using wild boostrapping to account for clustering at the clinic level (Cameron, Gelbrach, and Miller, 2008).p-values in parenthesis. *** p<0.01, ** p<0.05, * p<0.1
Anthropometrics (1) (2) (3)
BMI Upper-arm Waist
Eval*info -0.410 -0.125 -0.450
(0.268) (0.775) (0.605)
Eval*cook -0.912*** -0.596 -3.088***
(0.006) (0.108) (0.010)
Evaluation Dummy Yes Yes Yes
Individual FE Yes Yes Yes
Controls Yes Yes Yes
Mean control baseline 22.27 27.22 80.57
Observations 3,862 3,871 3,696
R-squared 0.032 0.093 0.114
Number of Ind. 1,964 1,962 1,908
Standard errors computed using wild boostrapping to account for clustering at the clinic level (Cameron, Gelbrach, and Miller, 2008).p-values in parenthesis. *** p<0.01, ** p<0.05, * p<0.1