negative rho values
Posted: 29 Jul 2025, 13:13
Hi
I am trying to do a route choice model. The data is not surveyed data but synthesized. The routes were generated based on directness measure assuming a unit demand. There were a total of 22 routes generated and the generated data were then grouped into 10 alternatives based on the normalized cdf of the excess directness measure and path size. The MNL was run on the groups assuming 22 individual and the choice being 1 to 10. However the results are not good. Since I didn't know what should be the path size for non chosen alternative (given that each of the 10 alternatives have the following proportions) I did the mnl on excess directness measure only.
Final identified cluster number Probability
1 0.181818
2 0.090909
3 0.272727
4 0.136364
5 0.045455
6 0.045455
7 0.045455
8 0.045455
9 0.045455
10 0.090909
the result was as follows:
Model name : MNL5847_final_paper
Model description : Simple MNL model on mode choice SP data
Model run at : 2025-07-29 16:48:08.974224
Estimation method : bgw
Model diagnosis : Relative function convergence
Optimisation diagnosis : Maximum found
hessian properties : Negative definite
maximum eigenvalue : -2.203768
reciprocal of condition number : 1
Number of individuals : 22
Number of rows in database : 22
Number of modelled outcomes : 22
Number of cores used : 1
Model without mixing
LL(start) : -50.66
LL at equal shares, LL(0) : -50.66
LL at observed shares, LL(C) : -45.64
LL(final) : -50.06
Rho-squared vs equal shares : 0.0117
Adj.Rho-squared vs equal shares : -0.008
Rho-squared vs observed shares : -0.0969
Adj.Rho-squared vs observed shares : 0.0783
AIC : 102.13
BIC : 103.22
Estimated parameters : 1
Time taken (hh:mm:ss) : 00:00:0.25
pre-estimation : 00:00:0.19
estimation : 00:00:0.03
post-estimation : 00:00:0.03
Iterations : 4
Unconstrained optimisation.
Estimates:
Estimate s.e. t.rat.(0) Rob.s.e. Rob.t.rat.(0)
asc_route1 0.0000 NA NA NA NA
d -0.7274 0.6736 -1.080 0.6498 -1.119
the rho -square values are negative. what should/can i do to improve them.
regards
Padma
I am trying to do a route choice model. The data is not surveyed data but synthesized. The routes were generated based on directness measure assuming a unit demand. There were a total of 22 routes generated and the generated data were then grouped into 10 alternatives based on the normalized cdf of the excess directness measure and path size. The MNL was run on the groups assuming 22 individual and the choice being 1 to 10. However the results are not good. Since I didn't know what should be the path size for non chosen alternative (given that each of the 10 alternatives have the following proportions) I did the mnl on excess directness measure only.
Final identified cluster number Probability
1 0.181818
2 0.090909
3 0.272727
4 0.136364
5 0.045455
6 0.045455
7 0.045455
8 0.045455
9 0.045455
10 0.090909
the result was as follows:
Model name : MNL5847_final_paper
Model description : Simple MNL model on mode choice SP data
Model run at : 2025-07-29 16:48:08.974224
Estimation method : bgw
Model diagnosis : Relative function convergence
Optimisation diagnosis : Maximum found
hessian properties : Negative definite
maximum eigenvalue : -2.203768
reciprocal of condition number : 1
Number of individuals : 22
Number of rows in database : 22
Number of modelled outcomes : 22
Number of cores used : 1
Model without mixing
LL(start) : -50.66
LL at equal shares, LL(0) : -50.66
LL at observed shares, LL(C) : -45.64
LL(final) : -50.06
Rho-squared vs equal shares : 0.0117
Adj.Rho-squared vs equal shares : -0.008
Rho-squared vs observed shares : -0.0969
Adj.Rho-squared vs observed shares : 0.0783
AIC : 102.13
BIC : 103.22
Estimated parameters : 1
Time taken (hh:mm:ss) : 00:00:0.25
pre-estimation : 00:00:0.19
estimation : 00:00:0.03
post-estimation : 00:00:0.03
Iterations : 4
Unconstrained optimisation.
Estimates:
Estimate s.e. t.rat.(0) Rob.s.e. Rob.t.rat.(0)
asc_route1 0.0000 NA NA NA NA
d -0.7274 0.6736 -1.080 0.6498 -1.119
the rho -square values are negative. what should/can i do to improve them.
regards
Padma