NYC-WTC_000187725

NYC 9/11 Public Portal Document

46 likelihood function which is maximized is of the form:® (24) Maximize ... ,4)5)) s

Finally, the values of the alternative specific dummies, I calculated so that predicted choice probabilities equal i! obse'rved relative frequencies of choice, This is done for each jkm;i given an arbitrary reference choice rxs;i: r ^Jkin; i (25) J

Selection of the solution Algorithm and calibration Results The main step in solving the model is the determination of the residential rents by housing type k and zone j. This is done by solving simultaneously for the R-jk’s the system of nonlinear simultaneous equations given by (IT). Then, shopping rents are computed from (12). To solve (9), two techniques have been developed and experimented with, and one has been adopted. The first technique used the simultaneous equation solver MINPACK which employs a hybrid Newton-Raphson type method based on the computation of the full Jacobian matrix of (11). MINPACK evaluates the Jacobian, not by analytical derivatives but, by numerically approximated

Q Because the data gives choices by j and m but not by housing type, it will be assumed in the likelihood function that the observed choices O-- are divided among the housing types in j according to the proportion of that housing type in that zone.

Q The values of the utilities are calibrated in such a way that the predicted number of commuters working in* i and choosing the peripheral super zone is equal to the observed number.

CITY_21MC100_006662277

NYC-WTC_000187777