This automated report is divided into three sections, one for each failure condition.
Failure Condition 1 analyses the relation between the parameter: Size Social Reference Network (mu_Rsocial), and the probability that the simulation resulted in no change, i.e. no reduction in the rate of FGM.
Failure Condition 2 analyses the relation between the parameter: Size Social Reference Network (mu_Rsocial), and the probability that the simulation resulted in negative spillover, i.e. the number of people abandoning at the end of the simulation is less than the number initially targeted by the intervention.
Failure Condition 3 analyses the relation between the parameter: Size Social Reference Network (mu_Rsocial), and the probability that the simulation only resulted in partial change, i.e. some people carried on practicing FGM at the end of the simulation.
The analysis used depends on whether the parameter is continuous or discrete
The first part of the analysis involves a technique called Monte Carlo Filtering (see Saltelli et al., who are cited in the thesis). In essence, this tests whether the cumulative distribution of the parameter varies between the situations where the failure scenario occurs and the situations where the failure scenario doesn’t occur. If it does, then this can be taken as an indication that the parameter has a marginal effect on the probability of the failure scenario. The test used to establish the difference is the KS-Test (with alpha set at 0.01)
The second part of the analysis involves a polynomial regression of the failure scenario on the parameter in question. This model is specified as
\[[Failure-Condition] = \beta_0 + \beta_1 [intervention-size] + \beta_2 [parameter] + \beta_3[parameter]^2 + \beta_4[parameter]^3 + \epsilon\] If one or more of the beta coefficients on the parameter is significant, this may indicate a relation between the parameter and the outcome.
The third part of the analysis is a visualization of the relation between the parameter value and the probability of the outcome, assuming an intervention of size 50%, based on the polynomial regression model.
When the parameter is discrete, the first part of the analysis is tabular. The simulation results are divided into groups based on intervention size, the groups are:
For each group, the discrete parameter is cross-tabulated with the failure condition, to give a probability that the failure condition occurred under each parameter value.
Next, a regression model is reported. This model has the following specification:
\[[Failure-Condition] = \beta_0 + \beta_1 [intervention-size] + \beta_2 [parameter] + \beta_2[parameter]\cdot[intervention-size]\] It reports the main effects of the discrete parameter (controlling for intervention size), as well as potential interactions with the intervention size.
Next, a plot shows a linear visualization of the relationship between intervention size and the probability of the failure condition. A new gradient is fit for each discrete value of the parameter, allowing the viewer to visualize the main effects of the parameters (i.e. changing intercept of the line) and possible interactions with intervention size (i.e. changing gradient of the line)
Dependent variable: | |
Failure Condition 1 | |
initial-target-proportion
|
-0.121*** |
(0.008) | |
Size Social Reference Network (mu_Rsocial)
|
0.006*** |
(0.001) | |
I(Size Social Reference Network (mu_Rsocial) 2)
|
-0.0001*** |
(0.00002) | |
I(Size Social Reference Network (mu_Rsocial) 3)
|
0.00000*** |
(0.00000) | |
Constant | 0.105*** |
(0.010) | |
Observations | 24,000 |
R2 | 0.015 |
Adjusted R2 | 0.014 |
Residual Std. Error | 0.336 (df = 23995) |
F Statistic | 88.544*** (df = 4; 23995) |
Note: | p<0.1; p<0.05; p<0.01 |
Dependent variable: | |
Failure Condition 2 | |
initial-target-proportion
|
0.049*** |
(0.011) | |
Size Social Reference Network (mu_Rsocial)
|
0.002* |
(0.001) | |
I(Size Social Reference Network (mu_Rsocial) 2)
|
-0.00004 |
(0.00003) | |
I(Size Social Reference Network (mu_Rsocial) 3)
|
0.00000 |
(0.00000) | |
Constant | 0.299*** |
(0.014) | |
Observations | 24,000 |
R2 | 0.001 |
Adjusted R2 | 0.001 |
Residual Std. Error | 0.475 (df = 23995) |
F Statistic | 6.368*** (df = 4; 23995) |
Note: | p<0.1; p<0.05; p<0.01 |
Dependent variable: | |
Failure Condition 3 | |
initial-target-proportion
|
-0.241*** |
(0.007) | |
Size Social Reference Network (mu_Rsocial)
|
-0.004*** |
(0.001) | |
I(Size Social Reference Network (mu_Rsocial) 2)
|
0.0001*** |
(0.00002) | |
I(Size Social Reference Network (mu_Rsocial) 3)
|
-0.00000*** |
(0.00000) | |
Constant | 1.046*** |
(0.009) | |
Observations | 24,000 |
R2 | 0.047 |
Adjusted R2 | 0.047 |
Residual Std. Error | 0.323 (df = 23995) |
F Statistic | 296.778*** (df = 4; 23995) |
Note: | p<0.1; p<0.05; p<0.01 |