Re-edited: problem in shaping unbalance panel data to run mixed multinomial regression

Question

I have followed 'mlogit' paper to prepare my unbalance panel data for multinomial (panel) regresssion, but I failed to run the regression. After reading some posts, I beleive I have found a good structure for the data, which is similar to this one . Nevertheless, the mlogit with panel data specification does not work. What I did wrong in the mlogit.data ? how can I solve it? Did I specify the mlogit model correctly?

My dataset is very similar to this one: http://dss.princeton.edu/training/Panel101.dta . Xs are independent variables arranged on a country-year basis (such as economic growth), while the dependent variable is opinion with 4 levels.

Panel <- Panel %>% select(country, year,opinion, x1, x2, x3)

Panel_multinomial <- mlogit.data(Panel, shape = 'wide', choice = "opinion", id ="country" , sep = "_")

this does not work:

 regression <- mlogit(opinion ~ x1+ x2+ x3, data = Panel_multinomial, 
        panel = TRUE, rpar = c(x1= "n", x2= "n", x3="n"), reflevel="1")

This does work:

regression <- mlogit(opinion ~ 1|x1+ x2+ x3, data = Panel_multinomial, reflevel="1")