Policy and Complex Systems
have beneficial effects for individuals with dementia and their caregivers on the days the individual attended the adult day center . These benefits include fewer behavior problems , better sleep , and decreased caregiver stress , lower cortisol levels ( indicating stress ), and less depression ( Gaugler et al ., 2003 ; Klein et al ., 2014 ; Zarit et al ., 2011 ; Zarit , Kim , Femia , Almeida , & Klein , 2014 ; Zarit , Stephens , Townsend , Greene , & Femia , 2003 ).
To address this problem , we recommend using agent-based modeling . The process of building a computational model requires quantifying the issues involved and combining theory , engineering approximations , and expert or at least the modeler ’ s opinions on how the world works and these should always be considered . The key presumptions of our model are presented above .
As demonstrated , agent-based modeling provides social service researchers and evaluators with a powerful tool for exploring and testing the impact of policy changes and assumptions as social service researchers use this modeling more frequently . There are strengths and limitations of this methodology that need to be borne in mind .
The reliability of these results requires much more than just analysis of the statistical significance of the results . The multiple runs conducted to support the significance testing only address the sensitivity of the results to the random numbers involved . In addition , we have conducted sensitivity analysis of three model design presumptions . We tested the presumption that a population of 100 older adults appropriate by trying populations of 50 and 200 . We varied the number of hours per day a caregiver was subject to stress . The default was 18 and we tested down to six and up to 24 hours per day . Finally , we tested the presumption relative weighting of ADLs and levels of dementia in contributing to the stress of the caregiver . The default presumption was that they both contributed equally to the level of stress of caregivers . We tested evaluations where ADLs were twice as important as the levels of dementia and vice versa . The results were not surprising except one . For a population of only 50 older adults , that was so few agents that one of the levels of adult day care relief ( 40 hours per week ) was statistically different at the 5 % level with 30 runs . We believe this was just a random event but in the process of doing the sensitivity analysis we did identify areas of the model that need research to improve the credibility of our model . And , of course , many more tests of the reliability of the model are possible and appropriate when simulation is used for regulatory and legislative purposes ( Hammond , 2015 ). However , for the purpose of presenting agent-based modeling as an approach to exploring policy options , we assert we have demonstrated the usefulness of this methodology .
Agent-based modeling allows us to explore and test policy options that are inconvenient , illegal , and immoral to perform with live human subjects . However , by representing humans in code , there are assumptions , approximations , and abstractions that have to be made and justified concerning the agents , their environment , and their interactions for the results to be creditable . The wise saying is that “ All models are wrong ; some are useful ” ( Epstein , 2008 ), because a simulation is not the actual phenomenon being studied .
Given the complicated nature of dementia caregiving , our future work will expand this model to incorporate other aspects of the care-giving decision-making process . We hope to include other family members in our model , and will begin to
39