问题
I am doing a discrete choice experiment on preferences for attributes of a hypothetical drug treating weight loss in my master thesis, and I need a little help.
My design is generic, and has 12 choice sets with three alternatives: Product A, Product B and Option out.
Somehow, I need to include option-out as an alternative specific constant, but it seems like I am doing something wrong here. I have 197 responses on 12 choice sets of three alternatives, hence 197*12*3 observations of choice = 7,092
> head(choice3, 12*3)
id choice_id mode.ids choice noadveff tab infreq_3 cost weightloss weightlosssq optout
1 x1 A 0 1 -1 -1 550 3.5 12.25 -1
1 x1 B 0 -1 1 1 90 6.0 36.00 -1
1 x1 C 1 0 0 0 0 0.0 0.00 1
1 x10 A 0 1 -1 1 50 6.0 36.00 -1
1 x10 B 0 -1 1 -1 165 3.5 12.25 -1
1 x10 C 1 0 0 0 0 0.0 0.00 1
1 x11 A 0 -1 -1 1 165 2.0 4.00 -1
1 x11 B 1 1 1 -1 90 3.5 12.25 -1
1 x11 C 0 0 0 0 0 0.0 0.00 -1
1 x12 A 0 -1 -1 1 550 6.0 36.00 -1
1 x12 B 0 1 1 -1 1000 2.0 4.00 -1
1 x12 C 1 0 0 0 0 0.0 0.00 1
1 x13 A 0 -1 -1 -1 90 6.0 36.00 -1
1 x13 B 0 1 1 1 1000 6.0 36.00 -1
1 x13 C 1 0 0 0 0 0.0 0.00 1
1 x2 A 0 -1 -1 -1 1000 6.0 36.00 -1
1 x2 B 0 1 1 1 300 2.0 4.00 -1
1 x2 C 1 0 0 0 0 0.0 0.00 1
1 x3 A 0 -1 -1 1 1000 6.0 36.00 -1
1 x3 B 1 1 1 -1 50 6.0 36.00 -1
1 x3 C 0 0 0 0 0 0.0 0.00 -1
1 x4 A 0 1 -1 1 165 3.5 12.25 -1
1 x4 B 0 -1 1 -1 550 2.0 4.00 -1
1 x4 C 1 0 0 0 0 0.0 0.00 1
1 x5 A 0 -1 -1 -1 550 2.0 4.00 -1
1 x5 B 1 1 1 1 50 6.0 36.00 -1
1 x5 C 0 0 0 0 0 0.0 0.00 -1
1 x6 A 0 1 -1 -1 300 6.0 36.00 -1
1 x6 B 0 -1 1 1 50 3.5 12.25 -1
1 x6 C 1 0 0 0 0 0.0 0.00 1
1 x8 A 0 -1 -1 1 300 3.5 12.25 -1
1 x8 B 1 1 1 -1 165 6.0 36.00 -1
1 x8 C 0 0 0 0 0 0.0 0.00 -1
1 x9 A 0 -1 1 -1 300 6.0 36.00 -1
1 x9 B 0 1 -1 1 90 2.0 4.00 -1
1 x9 C 1 0 0 0 0 0.0 0.00 1
I am using effects coding for my categorical variables (noadfeff, tab, infreq_3 and optout)
mode.ids indicate the alternatives, i.e. product A or B, or option-out (mode.ids == C)
The optout-variable has been coded with the following command,
choice2$optout <- ifelse(choice2$mode.ids == "C" & choice2$choice == 1, "1", "-1")
Using this data-set I set up the data for the mlogit-package
mlogit.all <- mlogit.data(choice2,
choice = "choice",
shape = c("long"),
id.var = "id",
alt.var = "mode.ids",
varying = c("cost", "cost.square", "noadvef", "tab", "infreq_3", "weightloss", "optout", "costsq", "weightlosssq"),
)
and run the following model
model.all <- mlogit(formula = choice ~ noadveff + tab + infreq_3 + cost + weightloss | optout | 0 ,
data = mlogit.all,
rpar = c(noadveff = 'n', tab = 'n', infreq_3 = 'n', weightloss = 'n', optout = 'u'),
R = 100,
halton = NA,
print.level = 0,
panel = TRUE
)
Which results in the following error.
Error in solve.default(H, g[!fixed]) :
Lapack routine dgesv: system is exactly singular: U[8,8] = 0
The similar model without optout, does not give any problems, so I must be doing something wrong with my optout-variable.
I hope you can see where the issue is? :)
Best, Henrik
回答1:
Sorry, I found the error myself - solution provided beneath,
choice2$optout <- ifelse(choice2$mode.ids == "C", "1", "-1")
Hence, we have a ASC equal to 1 for each alternative called C (i.e. the option out)
model.all <- mlogit(formula = choice ~ noadveff + tab + infreq_3 + cost + weightloss + optout | -1 | 0 ,
data = mlogit.all,
rpar = c(noadveff = 'n', tab = 'n', infreq_3 = 'n', weightloss = 'n'),
R = 100,
halton = NA,
print.level = 0,
panel = TRUE
)
After which the model will work, and generate an ASC for option-out'ers.
来源：https://stackoverflow.com/questions/31355145/including-opt-out-as-alternative-specific-constant-in-r-mlogit