log-likelihood calculation failed

[duxdfx]

Hi,

I am estimating a mixed logit model with 3 random coefficients and got the following error message:
"Error in apollo_estimate(apollo_beta, apollo_fixed, apollo_probabilities, :
Log-likelihood calculation fails at values close to the starting values!"
I tried to use the logit estimates for the starting values, but still didn't work. How should I proceed? I pasted the code below.

Thanks,
Sheldon

##----------------------------------
## estimation using Apollo
##----------------------------------
##following the example here: http://apollochoicemodelling.com/files/ ... ce_space.r
rm(list=ls())
setwd("Y:/Corn BE Yield Impact")

#install.packages("apollo")
library(apollo)

load("sample_logit_10000_forced_final.RData")
#head(data_sample)
#here "id" is the choice situation ("unit_id" + "crop_year"); "ids" is unit_id
colnames(data_sample)[which(colnames(data_sample)=="ids")] <- c("unit_id")

### Initialise code
apollo_initialise()

### Set core controls
apollo_control = list(
modelName = "MMNL_preference_space",
modelDescr = "Mixed logit model with demographics",
indivID = "unit_id",
mixing = TRUE,
nCores = 4,
outputDirectory = "output"
)

database = data_sample
head(database)

# ################################################################# #
#### DEFINE MODEL PARAMETERS ####
# ################################################################# #

### Vector of parameters, including any that are kept fixed in estimation; asc-alternative specific constant.
apollo_beta=c(asc_pln1 = 0, asc_pln2 = -0.70, asc_pln3 = 0.76, asc_pln4 = 2.08, asc_pln5 = 0.78,
asc_pln6 = -0.86, asc_pln7 = -4.11, asc_pln8 = -4.31, asc_pln9 = 1.38, asc_pln10 = 2.22,
asc_pln11 = 1.09, asc_pln12 = -0.95, asc_pln13 = -2.10, asc_pln14 = -1.21, asc_pln15 = -0.39,
asc_pln16 = 0.42, asc_pln17 = 1.40, asc_pln18 = 1.55, asc_pln19 = 0.16, asc_pln20 = -2.96, asc_pln21 = -5.925,
# b_prm = 0,
# b_sub = 0,
# b_lia = 0,
# asc_pln2_shift_acres = -0.00007, asc_pln3_shift_acres = 0.0003, asc_pln4_shift_acres = -0.0015,asc_pln5_shift_acres = -0.00078,asc_pln6_shift_acres = -0.0018,
# asc_pln7_shift_acres = -0.00077, asc_pln8_shift_acres = -0.0016,asc_pln9_shift_acres = 0.0016, asc_pln10_shift_acres = 0.0018, asc_pln11_shift_acres = 0.0014,
# asc_pln12_shift_acres = 0.0015, asc_pln13_shift_acres = 0.006, asc_pln14_shift_acres = 0.0019,asc_pln15_shift_acres = -0.001,asc_pln16_shift_acres = 0.0016,
# asc_pln17_shift_acres = 0.005, asc_pln18_shift_acres = 0.0009,asc_pln19_shift_acres = 0.000875,asc_pln20_shift_acres = 0.0014,asc_pln21_shift_acres = 0.0012,
# asc_pln2_shift_myield = -0.01, asc_pln3_shift_myield = -0.015,asc_pln4_shift_myield = -0.002,asc_pln5_shift_myield = 0.002,asc_pln6_shift_myield = 0.015,
# asc_pln7_shift_myield = 0.023, asc_pln8_shift_myield = 0.023,asc_pln9_shift_myield = -0.001,asc_pln10_shift_myield = -0.0006,asc_pln11_shift_myield = 0.007,
# asc_pln12_shift_myield = 0.013, asc_pln13_shift_myield = 0.016,asc_pln14_shift_myield = -0.0019,asc_pln15_shift_myield = -0.011,asc_pln16_shift_myield = -0.012,
# asc_pln17_shift_myield = -0.0012, asc_pln18_shift_myield = 0.00001,asc_pln19_shift_myield = 0.013,asc_pln20_shift_myield = 0.032,asc_pln21_shift_myield = 0.045,
# asc_pln2_shift_vyield = -0.028, asc_pln3_shift_vyield = 0.01,asc_pln4_shift_vyield = 0.0055,asc_pln5_shift_vyield = 0.016,asc_pln6_shift_vyield = 0.007,
# asc_pln7_shift_vyield = 0.037, asc_pln8_shift_vyield = 0.021,asc_pln9_shift_vyield = 0.012,asc_pln10_shift_vyield = 0.013,asc_pln11_shift_vyield = 0.019,
# asc_pln12_shift_vyield = 0.019, asc_pln13_shift_vyield = 0.03,asc_pln14_shift_vyield = 0.009,asc_pln15_shift_vyield = 0.026,asc_pln16_shift_vyield = 0.018,
# asc_pln17_shift_vyield = 0.013, asc_pln18_shift_vyield = 0.017,asc_pln19_shift_vyield = 0.018,asc_pln20_shift_vyield = 0.024,asc_pln21_shift_vyield = 0.03,
mu_log_b_prm = -0.004, sigma_log_b_prm = 0,
mu_log_b_sub = 0.004, sigma_log_b_sub = 0,
mu_log_b_lia = 0.0001, sigma_log_b_lia = 0)

### Vector with names (in quotes) of parameters to be kept fixed at their starting value in apollo_beta, use apollo_beta_fixed = c() if none
apollo_fixed = c("asc_pln1")

# ################################################################# #
#### DEFINE RANDOM COMPONENTS ####
# ################################################################# #

### Set parameters for generating draws
apollo_draws = list(
interDrawsType = "halton",
interNDraws = 500,
interUnifDraws = c(),
interNormDraws = c("draws_prm","draws_sub","draws_lia"),
intraDrawsType = "halton",
intraNDraws = 0,
intraUnifDraws = c(),
intraNormDraws = c()
)

### Create random parameters
apollo_randCoeff = function(apollo_beta, apollo_inputs){
randcoeff = list()

randcoeff[["b_prm"]] = -exp( mu_log_b_prm + sigma_log_b_prm * draws_prm )
randcoeff[["b_sub"]] = -exp( mu_log_b_sub + sigma_log_b_sub * draws_sub )
randcoeff[["b_lia"]] = -exp( mu_log_b_lia + sigma_log_b_lia * draws_lia )

return(randcoeff)
}

# ################################################################# #
#### GROUP AND VALIDATE INPUTS ####
# ################################################################# #

apollo_inputs = apollo_validateInputs()

# ################################################################# #
#### DEFINE MODEL AND LIKELIHOOD FUNCTION ####
# ################################################################# #

apollo_probabilities=function(apollo_beta, apollo_inputs, functionality="estimate"){

### Attach inputs and detach after function exit
apollo_attach(apollo_beta, apollo_inputs)
on.exit(apollo_detach(apollo_beta, apollo_inputs))

### Create list of probabilities P
P = list()

### Create alternative specific constants and coefficients using interactions with socio-demographics

# asc_pln2_value = asc_pln2 + asc_pln2_shift_acres * acres + asc_pln2_shift_myield * yield_mean + asc_pln2_shift_vyield * yield_var
# asc_pln3_value = asc_pln3 + asc_pln3_shift_acres * acres + asc_pln3_shift_myield * yield_mean + asc_pln3_shift_vyield * yield_var
# asc_pln4_value = asc_pln4 + asc_pln4_shift_acres * acres + asc_pln4_shift_myield * yield_mean + asc_pln4_shift_vyield * yield_var
# asc_pln5_value = asc_pln5 + asc_pln5_shift_acres * acres + asc_pln5_shift_myield * yield_mean + asc_pln5_shift_vyield * yield_var
# asc_pln6_value = asc_pln6 + asc_pln6_shift_acres * acres + asc_pln6_shift_myield * yield_mean + asc_pln6_shift_vyield * yield_var
# asc_pln7_value = asc_pln7 + asc_pln7_shift_acres * acres + asc_pln7_shift_myield * yield_mean + asc_pln7_shift_vyield * yield_var
# asc_pln8_value = asc_pln8 + asc_pln8_shift_acres * acres + asc_pln8_shift_myield * yield_mean + asc_pln8_shift_vyield * yield_var
# asc_pln9_value = asc_pln9 + asc_pln9_shift_acres * acres + asc_pln9_shift_myield * yield_mean + asc_pln9_shift_vyield * yield_var
# asc_pln10_value = asc_pln10 + asc_pln10_shift_acres * acres + asc_pln10_shift_myield * yield_mean + asc_pln10_shift_vyield * yield_var
# asc_pln11_value = asc_pln11 + asc_pln11_shift_acres * acres + asc_pln11_shift_myield * yield_mean + asc_pln11_shift_vyield * yield_var
# asc_pln12_value = asc_pln12 + asc_pln12_shift_acres * acres + asc_pln12_shift_myield * yield_mean + asc_pln12_shift_vyield * yield_var
# asc_pln13_value = asc_pln13 + asc_pln13_shift_acres * acres + asc_pln13_shift_myield * yield_mean + asc_pln13_shift_vyield * yield_var
# asc_pln14_value = asc_pln14 + asc_pln14_shift_acres * acres + asc_pln14_shift_myield * yield_mean + asc_pln14_shift_vyield * yield_var
# asc_pln15_value = asc_pln15 + asc_pln15_shift_acres * acres + asc_pln15_shift_myield * yield_mean + asc_pln15_shift_vyield * yield_var
# asc_pln16_value = asc_pln16 + asc_pln16_shift_acres * acres + asc_pln16_shift_myield * yield_mean + asc_pln16_shift_vyield * yield_var
# asc_pln17_value = asc_pln17 + asc_pln17_shift_acres * acres + asc_pln17_shift_myield * yield_mean + asc_pln17_shift_vyield * yield_var
# asc_pln18_value = asc_pln18 + asc_pln18_shift_acres * acres + asc_pln18_shift_myield * yield_mean + asc_pln18_shift_vyield * yield_var
# asc_pln19_value = asc_pln19 + asc_pln19_shift_acres * acres + asc_pln19_shift_myield * yield_mean + asc_pln19_shift_vyield * yield_var
# asc_pln20_value = asc_pln20 + asc_pln20_shift_acres * acres + asc_pln20_shift_myield * yield_mean + asc_pln20_shift_vyield * yield_var
# asc_pln21_value = asc_pln21 + asc_pln21_shift_acres * acres + asc_pln21_shift_myield * yield_mean + asc_pln21_shift_vyield * yield_var


### List of utilities: these must use the same names as in mnl_settings, order is irrelevant
V = list()
V[["pln1"]] = asc_pln1 + b_prm * prm1 + b_sub * sub1 + b_lia * lia1
V[["pln2"]] = asc_pln2 + b_prm * prm2 + b_sub * sub2 + b_lia * lia2
V[["pln3"]] = asc_pln3 + b_prm * prm3 + b_sub * sub3 + b_lia * lia3
V[["pln4"]] = asc_pln4 + b_prm * prm4 + b_sub * sub4 + b_lia * lia4
V[["pln5"]] = asc_pln5 + b_prm * prm5 + b_sub * sub5 + b_lia * lia5
V[["pln6"]] = asc_pln6 + b_prm * prm6 + b_sub * sub6 + b_lia * lia6
V[["pln7"]] = asc_pln7 + b_prm * prm7 + b_sub * sub7 + b_lia * lia7
V[["pln8"]] = asc_pln8 + b_prm * prm8 + b_sub * sub8 + b_lia * lia8
V[["pln9"]] = asc_pln9 + b_prm * prm9 + b_sub * sub9 + b_lia * lia9
V[["pln10"]] = asc_pln10 + b_prm * prm10 + b_sub * sub10 + b_lia * lia10
V[["pln11"]] = asc_pln11 + b_prm * prm11 + b_sub * sub11 + b_lia * lia11
V[["pln12"]] = asc_pln12 + b_prm * prm12 + b_sub * sub12 + b_lia * lia12
V[["pln13"]] = asc_pln13 + b_prm * prm13 + b_sub * sub13 + b_lia * lia13
V[["pln14"]] = asc_pln14 + b_prm * prm14 + b_sub * sub14 + b_lia * lia14
V[["pln15"]] = asc_pln15 + b_prm * prm15 + b_sub * sub15 + b_lia * lia15
V[["pln16"]] = asc_pln16 + b_prm * prm16 + b_sub * sub16 + b_lia * lia16
V[["pln17"]] = asc_pln17 + b_prm * prm17 + b_sub * sub17 + b_lia * lia17
V[["pln18"]] = asc_pln18 + b_prm * prm18 + b_sub * sub18 + b_lia * lia18
V[["pln19"]] = asc_pln19 + b_prm * prm19 + b_sub * sub19 + b_lia * lia19
V[["pln20"]] = asc_pln20 + b_prm * prm20 + b_sub * sub20 + b_lia * lia20
V[["pln21"]] = asc_pln21 + b_prm * prm21 + b_sub * sub21 + b_lia * lia21

### Define settings for MNL model component
mnl_settings = list(
alternatives = c(pln1=1, pln2=2, pln3=3, pln4=4, pln5=5, pln6=6, pln7=7, pln8=8, pln9=9, pln10=10, pln11=11, pln12=12,pln13=13, pln14=14, pln15=15,pln16=16, pln17=17, pln18=18,pln19=19, pln20=20, pln21=21),
avail = list(pln1=av1, pln2=av2, pln3=av3, pln4=av4, pln5=av5, pln6=av6, pln7=av7, pln8=av8,pln9=av9, pln10=av10, pln11=av11, pln12=av12,pln13=av13,pln14=av14,pln15=av15,pln16=av16,pln17=av17,pln18=av18,pln19=av19,pln20=av20,pln21=av21),
choiceVar = choice,
utilities = V
)

### Compute probabilities using MNL model
P[["model"]] = apollo_mnl(mnl_settings, functionality)

### Take product across observation for same individual
P = apollo_panelProd(P, apollo_inputs, functionality)

### Average across inter-individual draws
P = apollo_avgInterDraws(P, apollo_inputs, functionality)

### Prepare and return outputs of function
P = apollo_prepareProb(P, apollo_inputs, functionality)
return(P)
}

# ################################################################# #
#### MODEL ESTIMATION ####
# ################################################################# #

model = apollo_estimate(apollo_beta, apollo_fixed, apollo_probabilities, apollo_inputs)

# ################################################################# #
#### MODEL OUTPUTS ####
# ################################################################# #

# ----------------------------------------------------------------- #
#---- FORMATTED OUTPUT (TO SCREEN) ----
# ----------------------------------------------------------------- #

apollo_modelOutput(model)

# ----------------------------------------------------------------- #
#---- FORMATTED OUTPUT (TO FILE, using model name) ----
# ----------------------------------------------------------------- #

apollo_saveOutput(model)

[stephanehess]

Hi

this is quite a common starting values issue with Lognormal distributions - see the discussion at http://www.apollochoicemodelling.com/faq.html

Remember that the lognormal uses an exponential of a Normal, so if you are using MNL point values as the starting values for the means, you need to take the logs of those first

Stephane