#xFit/Error: cannot find valid starting values: please specify some

[anup545587]

Hi,
datimp1 <- neImpute(Srhrecodedop ~ education_lvls+smokingstatusnew_bmi_binnedrpa3lvls+STARTALD,
family = binomial("logit"), nMed = 3, data = new)
expFit <- glm( education_lvls~ STARTALD, family = binomial("logit"), data = new)
nemodd1 <- neModel(Srhrecodedop ~ education_lvls0+education_lvls1 + STARTALD,
family = binomial("logit"), expData = datimp1,xFit = expFit, se="robust")
Error: cannot find valid starting values: please specify some

education_lvls is a categorical variable. cannot figure out the source of error.

[jmpsteen]

Thanks for your interest in medflex! My guess is that this error is not specific for medflex but originates from the function that is called internally (glm). Which of the three above commands produces the actual error?

[kristyrobledo]

Loving the package - thankyou @jmpsteen !

I've also encountered the "Error: cannot find valid starting values: please specify some" problem. I've provided a reproducible example (hopefully!) below. The analysis works for performing a glm (see the last check, the are warnings but the model has converged). Are the starting values from neModel being passed to the glm model?

library(palmerpenguins)
library(medflex)
#> Loading required package: multcomp
#> Loading required package: mvtnorm
#> Loading required package: survival
#> Loading required package: TH.data
#> Loading required package: MASS
#>
#> Attaching package: 'TH.data'
#> The following object is masked from 'package:MASS':
#>
#> geyser
#> medflex 0.6-7: Flexible Mediation Analysis Using Natural Effect Models
#> Please report bugs here: github.com/jmpsteen/medflex/issues
library(tidyverse)
library(reprex)

create a fake dataset to show example of error

glimpse(penguins)
#> Rows: 344
#> Columns: 8
#> $ species Adelie, Adelie, Adelie, Adelie, Adelie, Adelie, Adel~
#> $ island Torgersen, Torgersen, Torgersen, Torgersen, Torgerse~
#> $ bill_length_mm 39.1, 39.5, 40.3, NA, 36.7, 39.3, 38.9, 39.2, 34.1, ~
#> $ bill_depth_mm 18.7, 17.4, 18.0, NA, 19.3, 20.6, 17.8, 19.6, 18.1, ~
#> $ flipper_length_mm 181, 186, 195, NA, 193, 190, 181, 195, 193, 190, 186~
#> $ body_mass_g 3750, 3800, 3250, NA, 3450, 3650, 3625, 4675, 3475, ~
#> $ sex male, female, female, NA, female, male, female, male~
#> $ year 2007, 2007, 2007, 2007, 2007, 2007, 2007, 2007, 2007~

penguins %>%
filter(species!="Gentoo") %>%
mutate(treat = as.numeric(rbernoulli(n=220, 0.5))) ->df

summary(df)
#> species island bill_length_mm bill_depth_mm
#> Adelie :152 Biscoe : 44 Min. :32.10 Min. :15.50
#> Chinstrap: 68 Dream :124 1st Qu.:37.75 1st Qu.:17.50
#> Gentoo : 0 Torgersen: 52 Median :40.60 Median :18.40
#> Mean :41.91 Mean :18.37
#> 3rd Qu.:45.95 3rd Qu.:19.10
#> Max. :58.00 Max. :21.50
#> NA's :1 NA's :1
#> flipper_length_mm body_mass_g sex year treat
#> Min. :172.0 Min. :2700 female:107 Min. :2007 Min. :0.0000
#> 1st Qu.:187.0 1st Qu.:3400 male :107 1st Qu.:2007 1st Qu.:0.0000
#> Median :191.0 Median :3700 NA's : 6 Median :2008 Median :0.0000
#> Mean :191.8 Mean :3711 Mean :2008 Mean :0.4591
#> 3rd Qu.:196.0 3rd Qu.:3988 3rd Qu.:2009 3rd Qu.:1.0000
#> Max. :212.0 Max. :4800 Max. :2009 Max. :1.0000
#> NA's :1 NA's :1

##fit multiple mediator model, using binary regression (log link) to get RR not OR
rr.m <- glm(factor(species) ~factor(treat)+
bill_length_mm +bill_depth_mm +
flipper_length_mm,
data=df,
family = binomial(link="log"),
start=c(log(0.15), 0,0,0,0),
control=list(maxit=10000))
#> Warning: step size truncated due to divergence
#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds
#> Warning: glm.fit: algorithm stopped at boundary value
#> Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred
expData<-neImpute(rr.m,
nMed = 3,
data=df)
#works, everything OK

#fit NE model
neMod<-neModel(factor(species) ~ treat0+treat1+
flipper_length_mm,
family = binomial(link = "log"),
start=c(log(0.15), 0,0,0),
control=list(maxit=10000),
se="robust",
expData=expData)
#> Error: no valid set of coefficients has been found: please supply starting values
#errors: "Error: no valid set of coefficients has been found: please supply starting values"

#check glm model
check<-glm(factor(species) ~ treat0+treat1+
flipper_length_mm,
family = binomial(link = "log"),
start=c(log(0.15), 0,0,0),
control=list(maxit=10000),
data=expData)
#> Warning: step size truncated due to divergence
#> Warning: step size truncated due to divergence

#> Warning: step size truncated due to divergence
#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds

#> Warning: step size truncated: out of bounds
#> Warning: glm.fit: algorithm stopped at boundary value
#> Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

summary(check)
#>
#> Call:
#> glm(formula = factor(species) ~ treat0 + treat1 + flipper_length_mm,
#> family = binomial(link = "log"), data = expData, start = c(log(0.15),
#> 0, 0, 0), control = list(maxit = 10000))
#>
#> Deviance Residuals:
#> Min 1Q Median 3Q Max
#> -3.0634 0.0896 0.1046 0.1225 0.1461
#>
#> Coefficients:
#> Estimate Std. Error z value Pr(>|z|)
#> (Intercept) 0.0182275 0.0414674 0.440 0.660
#> treat01 -0.0032772 0.0066597 -0.492 0.623
#> treat11 0.0027601 0.0065109 0.424 0.672
#> flipper_length_mm -0.0001220 0.0002202 -0.554 0.580
#>
#> (Dispersion parameter for binomial family taken to be 1)
#>
#> Null deviance: 14.162 on 437 degrees of freedom
#> Residual deviance: 14.218 on 434 degrees of freedom
#> (2 observations deleted due to missingness)
#> AIC: 22.218
#>
#> Number of Fisher Scoring iterations: 47
check$converged
#> [1] TRUE

warnings are present, but model converges and looks ok.....

are the extra settings (start values above) being passed to the GLM?

Created on 2022-05-18 by the reprex package (v2.0.1)