• Stochastic Simulation Approach
    • Each building has two recovery states (recovered/not recovered)
    • Two possibilities
      • We have recovery data for all damaged buildings
        • Simulate recovery at the building scale and then aggregate to census block or study region scales
      • We have recovery data for some of the damaged buildings
        • Use bootstrap to sample buildings and model recovery at the census block and study-region scales
    • State transition modeled using Poisson distribution
      • Time-based model
        • Regression performed on data from field survey and building department using the recovery time as the dependent variable. Various explanatory/independent variables (census data, building damage, building value, building age etc.)
          • Different types of statistical models used for regression including machine learning algorithms
        • Use inverse method and monte carlo simulation to generate the recovery time
      • State-based model
        • Regression performed on data from field survey and building department using the rate parameter from the exponential distribution as the dependent variable. Various explanatory/independent variables (census data, building damage, building value, building age etc.)
          • Different types of statistical models used for regression including machine learning algorithms
        • At any given point in time, we can compute the probability of recovery given the rate parameter from the exponential distribution
        • Use monte carlo to simulate the state of the building at any given point in time
  • Statistical Model
    • Each building has two recovery states (recovered/not recovered)
    • Two possibilities
      • We have recovery data for all damaged buildings
        • Simulate recovery at the building scale and then aggregate to census block or study region scales
      • We have recovery data for some of the damaged buildings
        • Use bootstrap to sample buildings and model recovery at the census block and study-region scales
    • Logistic regression performed on data from field survey and building department using the probability of recovery the dependent variable. Various explanatory/independent variables (time, census data, building damage, building value, building age etc.)
      • Different types of statistical models used for regression including machine learning algorithms
        • Use inverse method and monte carlo simulation to generate the recovery time
    • Use monte carlo to simulate the state of the building at any given point in time