Policy and Complex Systems
intervention were used to populate the initial state of the simulation model and to validate the results of the simulation model . Data included demographics , experience , and content knowledge of teachers , classroom observations , student pre-post SLIDER test scores , class sizes , and support for intervention of the teachers and administration . Other data were acquired through the Georgia Department of Education . Any remaining data needed for these models were acquired through surveys given to the SLIDER team at GT and the teachers and administration at the schools involved in the intervention . The simulation model was built using the object-oriented programming language C # in Microsoft Visual Studio 2010 and run on a personal computer . The stochastic models were run 1000 times and averaged to approximate a deterministic output , which is needed for sensitivity analysis and variable screening methods . Complete sensitivity analysis results for the SLIDER case study can be found in ( Mital , 2015 ), where the model was found not to be unusually sensitive to the selection of any one parameter .
3.4 Model Validation
The verification and validation steps implemented during the different phases are discussed briefly here . Some of these were compiled by Sargent in his work on validation and verification of simulation models ( Sargent , 2004 ).
Conceptual model and face validation : Ten SMEs were used to test whether the model and its behavior were conceptually logical and whether the model ’ s input – output relationships were reasonable . They examined the model for completeness , consistency , coherence , and correctness as described in the framework proposed by Pace ( 2000 ).
Data validation : The changes in the attributes of the agents were compared to the data collected with respect to these attributes .
Comparison to Other Models : The model results obtained are consistent with educational research studies discussed in the literature review .
Parameter variability — Sensitivity analysis : The model was run under different sets of parameter and input conditions and model outputs were analyzed .
4 . Analysis and Insights
For brevity , individual attribute
simulation results are not presented here , but can be found in ( Mital , 2015 ). Figures 3 and 4 depict the changes in the gaps for Rural School and Suburban School , respectively , with the blue line representing the simulation results , the red line representing reality ( as determined from survey data ), and the green box representing the acceptable or sustainability boundaries as modeled . As can be seen in Figure 3 , Rural School starts as a “ higher risk ” school in that the Ca gap is well outside of the acceptable zone , and the Cu gap is on the edge of acceptability . However , Rural School consistently moves toward the acceptable zone , and by the end of year 4 , Rural School is within the acceptable zone or close to it for all three gaps . This is consistent with the survey data and reality — Rural School has maintained many aspects of the intervention since the end of the SLIDER grant . On the other hand , Suburban School , which started in a “ low risk ” state relative to Rural School , continues to worsen during the grant period ,
81