Parameter does not influence the log-likelihood of your model!

[Dian3426]

Hi Stephane and David,

Hope you are doing great!

I am trying to use the Apollo package (0.2.9) to run the MNL_RP model.

Here is the error message I got from R (version 4.2.0; RStudio2022.02.1):
"Error in apollo_estimateHB(apollo_beta, apollo_fixed, apollo_probabilities, :
SPECIFICATION ISSUE - Parameter b_disc.d2 does not influence the log-likelihood of your model!"


Below is the code I used to run the analysis.

# Import

dce = read.csv(file=paste0(PATHDAT,'dce_wide.csv'), header = T, stringsAsFactors = F)

load(paste0(PATHOBJ,'List of model specifications.RData')) # l.spe
load(paste0(PATHOBJ,'List of attributes.RData')) # l.att


# Manage --------------------------------------------------

###
MODL = 'RPL_effect'
OBJT = 'RPL_effect - Model estimates_child'
SAMP = 'Overall'

### META Variables
N = 'record'
Ta = 'SP_task'
Y = "choice"
nJ = 2

###
nfac = 1000
nBURN = 300*nfac
nCONV = 30*nfac
nSKIP = 10
SEED = 12345


# Specification -----------------------------------------------------------

lD = as.list(NULL)
for (i in 1) {
###
D = dce

###
lA = l.att
S0 = l.spe[['eff0']]
S1 = l.spe[['eff1']]

### Model specification
lX = paste0("b_",S1)
lparam = rep(0,length(lX))
names(lparam) = lX

###
hbd = NULL
hbd = rep('N',length(lX))
names(hbd) = lX

###
l.uti = as.list(NULL)
l.uti = lapply(1:nJ, function(x){
a = paste0(lX,"*alt",x,'_',sub('b\\_','',lX),collapse=" +")
return(a)})

###
nN = length(unique(D[,N]))
nY = nrow(D)
nP = 2*length(hbd)
LL0 = nY*log(1/nJ)
nD = nrow(D)

###
mID = D[!duplicated(D[,'ID']),c('ID',N)]

###
lk = as.list(NULL)
lk[['lA']] = lA
lk[['S0']] = S0
lk[['S1']] = S1
lk[['D']] = D
lk[['nN']] = nN
lk[['nY']] = nY
lk[['nP']] = nP
lk[['LL0']] = LL0
lk[['hbd']] = hbd
lk[['lparam']] = lparam
lk[['l.uti']] = l.uti
lk[['mID']] = mID
lk[['nD']] = nD

lD[] = lk}

# Estimation ------------------------------------------------------------------------

###
l.res = as.list(NULL)

IDXTOT = length(lD)

for (i in IDXTOT){
###
lparam = lD[][['lparam']]
l.uti = lD[][['l.uti']]
hbd = lD[][['hbd']]
database = lD[][['D']]
nD = lD[][['nD']]
nY = lD[][['nY']]
nN = lD[][['nN']]
nP = lD[][['nP']]
LL0 = lD[][['LL0']]
mID = lD[[i]][['mID']]
S0 = lD[[i]][['S0']]
S1 = lD[[i]][['S1']]
lA = lD[[i]][['lA']]

### Initialise code
apollo_initialise()

### Set core controls
apollo_control = list(
modelName =MODL,
modelDescr =MODL,
indivID ='ID',
HB =T)

### Declare parameters
apollo_beta = lparam
apollo_fixed = c()

###
apollo_HB = list(
hbDist =hbd,
gINFOSKIP =nSKIP,
gNCREP =nBURN,
gNEREP =nCONV,
gSeed =SEED,
verbose =F,
nodiagnostics=T,
gFULLCV =F)

### VALIDATE INPUTS
apollo_inputs = apollo_validateInputs()

### DEFINE MODEL
apollo_probabilities = function(apollo_beta, apollo_inputs, functionality="estimate"){
###
apollo_attach(apollo_beta, apollo_inputs)
on.exit(apollo_detach(apollo_beta, apollo_inputs))

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

### Likelihood of choices
V = list()
V[['alt1']] = (b_adpro.d2 * alt1_adpro.d2 +
b_inj.d2 * alt1_inj.d2 + b_inj.d3 * alt1_inj.d3 +
b_disc.d2 * alt1_disc.d2 + b_disc.d3 * alt1_disc.d3 +
b_dos.d2 * alt1_dos.d2 + b_dos.d3 * alt1_dos.d3 +
b_visits.d2 * alt1_visits.d2 + b_visits.d3 * alt1_visits.d3)
V[['alt2']] = (b_adpro.d2 * alt2_adpro.d2 +
b_inj.d2 * alt2_inj.d2 + b_inj.d3 * alt2_inj.d3 +
b_disc.d2 * alt2_disc.d2 + b_disc.d3 * alt2_disc.d3 +
b_dos.d2 * alt2_dos.d2 + b_dos.d3 * alt2_dos.d3 +
b_visits.d2 * alt2_visits.d2 + b_visits.d3 * alt2_visits.d3)

### Define MNL settings
mnl_settings = list(
alternatives=c(alt1=1, alt2=2),
avail = list(alt1=1, alt2=1),
choiceVar = choice,
V = V)

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

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

### Estimate model
fit = NULL
fit = apollo_estimate(apollo_beta, apollo_fixed, apollo_probabilities, apollo_inputs)

.......



Do you have any suggestions to solve this error? Thank you very much in advance for your help

Best regards,
Dian

[stephanehess]

Hi

this would happen if there are no differences between the alternatives in the associated attribute, i.e. alt1_disc.d2 and alt2_disc.d2

Can you confirm that in your data, these do differ?

Stephane

[gabystephanie]

Hi Stephane and David,

Hope you are doing great!
I have been struggling with this issue for a while, and your help would be greatly appreciated.
I am trying to use the Apollo package to estimate the prior from my pilot test, but I keep getting this error message "Parameter does not influence the log-likelihood of your model!".

Below is the code I used to run the analysis.

apollo_initialise()
database<-data_long1
apollo_control=list(modelName = "dce pilot", modelDescr = "MNL model", indivID = "ID")
str(database)

apollo_beta<-c(asc_app1 = 0, asc_app2 = 0, b_ht_bud = 0, b_ht_stan = 0, b_ht_lux = 0, b_tc_low = 0, b_tc_med = 0, b_tc_high = 0, b_ev_low = 0, b_ev_med = 0, b_ev_high = 0)
apollo_fixed<-c("asc_app2", "b_ht_bud", "b_tc_low", "b_ev_low")
apollo_control
apollo_readBeta
apollo_inputs<-apollo_validateInputs()

apollo_probabilities<-function(apollo_beta,apollo_inputs,functionality="estimate"){
apollo_attach(apollo_beta,apollo_inputs)
on.exit(apollo_detach(apollo_beta,apollo_inputs))
P<-list()
V<-list()
V[['alt1']]=(asc_app1+b_ht_bud*(ht_bud==1)+b_ht_stan*(ht_stan==2)+b_ht_lux*(ht_lux==3)+b_tc_low*(tc_low==1)+b_tc_med*(tc_med==2)+b_tc_high*(tc_high==3)+b_ev_low*(ev_low==1)+b_ev_med*(ev_med==2)+b_ev_high*(ev_high==3))
V[['alt2']]=(asc_app2+b_ht_bud*(ht_bud==1)+b_ht_stan*(ht_stan==2)+b_ht_lux*(ht_lux==3)+b_tc_low*(tc_low==1)+b_tc_med*(tc_med==2)+b_tc_high*(tc_high==3)+b_ev_low*(ev_low==1)+b_ev_med*(ev_med==2)+b_ev_high*(ev_high==3))
mnl_settings<-list(alternatives=c(alt1=1,alt2=2), avail=list(alt1=1,alt2=1), choiceVar=choice, V=V)
P[['model']] <- apollo_mnl(mnl_settings, functionality)
P<- apollo_panelProd(P, apollo_inputs, functionality)
P<- apollo_prepareProb(P, apollo_inputs, functionality)
return(P)
}

model<-apollo_estimate(apollo_beta,apollo_fixed,apollo_probabilities,apollo_inputs)
apollo_modelOutput(model)

Preparing user-defined functions.

Testing likelihood function...

Overview of choices for MNL model component :
alt1 alt2
Times available 456 456
Times chosen 228 228
Percentage chosen overall 50 50
Percentage chosen when available 50 50


Pre-processing likelihood function...

Testing influence of parametersError in apollo_estimate(apollo_beta, apollo_fixed, apollo_probabilities, :
SPECIFICATION ISSUE - Parameters b_ht_stan, b_ht_lux, b_tc_med, b_tc_high, b_ev_med, b_ev_high do not influence the log-likelihood of your model!
------------------------------------------------------------------------------------------------------------------------
Do you have any suggestions to solve this error? Thank you very much in advance for your help!

Best regards,
Gaby

[stephanehess]

Hi Gaby

only differences in utility matter. If you include e.g.

Code: Select all

b_ht_stan*(ht_stan==2)

in both utilities, then it will simply drop out

Stephane

[mboer]

Dear dr. Hess,

I'm struggling with fitting a latent class MNL model. Please find my code below.

I keep getting the following error message: Error in apollo_estimate(apollo_beta, apollo_fixed, apollo_probabilities, :
SPECIFICATION ISSUE - Parameters b_reistijd_1_a, b_reistijd_2_a, b_frequentie_2_a, b_frequentie_3_a, b_frequentie_4_a, b_frequentie_5_a, b_reistijd_1_b, b_reistijd_2_b, b_frequentie_2_b, b_frequentie_3_b, b_frequentie_4_b, b_frequentie_5_b do not influence the log-likelihood of your model!

I have read your reply in this post and checked whether 'frequentie' differs across choice option 1 and 2, and it does, so I don't think that is the issue. For some reason, when I remove the +-sign in the equation for 'alt2' before defining the frequentie-dummy variables (see comment after @@), the model runs, but then I am afraid that it does not use all data for 'alt2'. Could you perhaps see what goes wrong? Any help is much appreciated.

Best,

M.

apollo_initialise()
apollo_control <- list(
modelName = "LCA_pilot_2classes",
modelDescr = "Latent class analysis pilot study",
indivID = "id",
outputDirectory = "Results/Pilot/LCA"
)

database <- readRDS("Data/Pilot/database_pilot.rds")

# Initialize betas
apollo_beta <- c(
b_reistijd_1_a = 0,
b_reistijd_2_a = 0,
b_reistijd_3_a = 0,
b_frequentie_1_a = 0,
b_frequentie_2_a = 0,
b_frequentie_3_a = 0,
b_frequentie_4_a = 0,
b_frequentie_5_a = 0,
b_reistijd_1_b = 0,
b_reistijd_2_b = 0,
b_reistijd_3_b = 0,
b_frequentie_1_b = 0,
b_frequentie_2_b = 0,
b_frequentie_3_b = 0,
b_frequentie_4_b = 0,
b_frequentie_5_b = 0,
delta_a = 0,
delta_b = 0,
b_asc_1 = 0,
b_asc_2 = 0,
b_out = 0
)

# Refcat
apollo_fixed <- c(
"b_reistijd_3_a", "b_frequentie_1_a",
"b_reistijd_3_b", "b_frequentie_1_b", "delta_a", "b_asc_1")

apollo_lcPars = function(apollo_beta, apollo_inputs) {
lcpars = list()
lcpars[["b_reistijd_1"]] = list(b_reistijd_1_a, b_reistijd_1_b)
lcpars[["b_reistijd_2"]] = list(b_reistijd_2_a, b_reistijd_2_b)
lcpars[["b_reistijd_3"]] = list(b_reistijd_3_a, b_reistijd_3_b)
lcpars[["b_frequentie_1"]] = list(b_frequentie_1_a, b_frequentie_1_b)
lcpars[["b_frequentie_2"]] = list(b_frequentie_2_a, b_frequentie_2_b)
lcpars[["b_frequentie_3"]] = list(b_frequentie_3_a, b_frequentie_3_b)
lcpars[["b_frequentie_4"]] = list(b_frequentie_4_a, b_frequentie_4_b)
lcpars[["b_frequentie_5"]] = list(b_frequentie_5_a, b_frequentie_5_b)

#Utilities of class allocation model
V = list()
V[["class_a"]] = delta_a
V[["class_b"]] = delta_b

#Settings for class allocation models
classAlloc_settings = list(
classes = c(class_a = 1, class_b = 2),
utilities = V
)
lcpars[["pi_values"]] = apollo_classAlloc(classAlloc_settings)
return(lcpars)

}

# Check inputs
apollo_inputs <- apollo_validateInputs()

# Define model
apollo_probabilities <- function(apollo_beta, apollo_inputs, functionality = "estimate"){
apollo_attach(apollo_beta, apollo_inputs)
on.exit(apollo_detach(apollo_beta, apollo_inputs))

# List for probabilities
P <- list()

# MNL settings
mnl_settings <- list(
alternatives = c(alt1 = 1, alt2 = 2, alt3 = 3),
avail = list(alt1 = 1, alt2 = 1, alt3 = 1),
choiceVar = keuze
)

for(s in 1:length(pi_values)){
# List for utilities
V <- list()
# Define formulas
V[['alt1']] = b_asc_1 +
b_reistijd_1[[s]]*(reistijd_1 == 1) +
b_reistijd_2[[s]]*(reistijd_1 == 2) +
b_reistijd_3[[s]]*(reistijd_1 == 3) +
b_frequentie_1[[s]]*(frequentie_1 == 1) +
b_frequentie_2[[s]]*(frequentie_1 == 2) +
b_frequentie_3[[s]]*(frequentie_1 == 3) +
b_frequentie_4[[s]]*(frequentie_1 == 4) +
b_frequentie_5[[s]]*(frequentie_1 == 5)
V[['alt2']] = b_asc_2 +
b_reistijd_1[[s]]*(reistijd_2 == 1) +
b_reistijd_2[[s]]*(reistijd_2 == 2) +
b_reistijd_3[[s]]*(reistijd_2 == 3) + # @@If I remove the + here, then it runs
b_frequentie_1[[s]]*(frequentie_2 == 1) +
b_frequentie_2[[s]]*(frequentie_2 == 2) +
b_frequentie_3[[s]]*(frequentie_2 == 3) +
b_frequentie_4[[s]]*(frequentie_2 == 4) +
b_frequentie_5[[s]]*(frequentie_2 == 5)
V[['alt3']] = b_out

mnl_settings$utilities = V

### Compute within-class choice probabilities using MNL model
P[[paste0("Class_",s)]] = apollo_mnl(mnl_settings, functionality)

### Take product across observation for same individual
P[[paste0("Class_",s)]] = apollo_panelProd(P[[paste0("Class_",s)]], apollo_inputs ,functionality)
}

### Compute latent class model probabilities
lc_settings = list(inClassProb = P, classProb=pi_values)
P[["model"]] = apollo_lc(lc_settings, 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)

[stephanehess]

Hi

well, it seems like most of your parameters are not having an effect.

First, did you estimate an MNL model and can you confirm that that worked fine?
Second, what version of Apollo are you using?

Stephane

[mboer]

Dear Stephane,

Thanks for your fast reply.

Yes, I first fitted an MNL model, which worked just fine, although in the ('1-class') MNL the parameters for 'frequentie' were either not in the expected direction or not significant.

Regarding your second question: I'm using Apollo version 0.3.4.

M.

[stephanehess]

Hi

try to update to the newest version of Apollo and also use different starting values in the two classes. If that does not help, then please let me know

Stephane

[mboer]

Dear Stephane,

I have updated Apollo to version 0.3.5 now. When I run the model with the script as specified as earlier, I get this message:

Error in apollo_lc(c(lc_settings, componentName2 = "model"), apollo_inputs, :
INPUT ISSUE - At your starting values, the probabilities are the same across some of the classes.Please use different starting values across classes. If you still wish to run your model in the current form, please set noValidation=TRUE in apollo_control.

I have tried to search for starting values using
apollo_beta=apollo_searchStart(apollo_beta, apollo_fixed,apollo_probabilities, apollo_inputs)

but then I get the following message:
Testing probability function (apollo_probabilities)
Apollo found a model component of type classAlloc without a componentName. The name was set to "classAlloc" by default.
INFORMATION: Setting "avail" is missing, so full availability is assumed.
Apollo found a model component of type MNL without a componentName. The name was set to "Class_1" by default.
Apollo found a model component of type MNL without a componentName. The name was set to "Class_2" by default.
Error in apollo_lc(c(lc_settings, componentName2 = "model"), apollo_inputs, :
INPUT ISSUE - At your starting values, the probabilities are the same across some of the classes.Please use different starting values across classes. If you still wish to run your model in the current form, please set noValidation=TRUE in apollo_control.

Can you see what is going wrong? Perhaps I made a coding error?

[stephanehess]

Hi

just change the starting values in apollo_beta so they are not exactly the same for the different classes

Stephane