*********************************************************************************

*				MODELLI PARAMETRICI DI IMPUTAZIONE UNIVARIATA

*********************************************************************************
keep maxgrip mergeid country wave gender age_int hhsize interview adl
drop if interview!=1 | age_int<0 | age_int==.
  
  recode maxgrip(min/-1=.)
  mean maxgrip
  recode maxgrip(.=34.2)
  
  recode adl(min/-1=.)
  *abbiamo circa 6000 valori mancanti, circa il 9%
  misstable summarize
  tab adl,mi
tabstat adl, st(mean sd var)
  
  
  *********************
  mi set mlong 
mi register imputed adl
mi passive: gen age2=age_int*age_int
mi register regular  mergeid country wave gender age_int hhsize maxgrip


**********************************************************************************
*			Poisson imputation model
			************************
* For a count variable = N. of limitations in daily living
* mean = variance
**********************************************************************************
mi impute poisson adl maxgrip i.country wave age_int age2 hhsize, by(gender) add(20) rseed(2232) noisily

  midiagplots adl
  midiagplots adl, plottype(hist) separate
  mi xeq 0 1 10 20: summarize adl
mi xeq 0 1 10 20: tab adl gender,col


**********************************************************************************
*			Negative binomial imputation model
			**********************************
* For a count variable
* overdispersion: variance > mean (in this case: mean= .227 sd= .842
**********************************************************************************
mi impute nbreg adl maxgrip i.country wave gender age_int age2 hhsize, by(gender) replace rseed(2232) noisily

  midiagplots adl
  midiagplots adl, plottype(hist) separate
  mi xeq 0 1 10 20: summarize adl
mi xeq 0 1 10 20: tab adl


**********************************************************************************
*				Ordered logit imputation model
				******************************
* For ordinal variables
* Proportional odds (the effect is proportional across cutoffs)
**********************************************************************************
mi impute ologit adl maxgrip i.country wave gender age_int age2 hhsize, by(gender) replace rseed(2232) noisily

  midiagplots adl
  midiagplots adl, plottype(hist) separate
  mi xeq 0 1 10 20: summarize adl
mi xeq 0 1 10 20: tab adl



**********************************************************************************
*					Logistic imputation model
					*************************
* For binary variables
**********************************************************************************
recode adl(1/6=1),gen(adl2)
mi register imputed adl2
 
mi impute logit adl2 maxgrip i.country wave gender age_int age2 hhsize, conditional(if gender==1) replace rseed(2232) noisily
mi impute logit adl2 maxgrip i.country wave gender age_int age2 hhsize, conditional(if gender==2) replace rseed(2232) noisily

  midiagplots adl2
  midiagplots adl2, plottype(hist) separate
  mi xeq 0 1 10 20: summarize adl2
mi xeq 0 1 10 20: tab adl2



**********************************************************************************
*				Multinomial logistic imputation model
				*************************************
*
**********************************************************************************
recode adl(1/3=1) (4/6=2),gen(adl3)
mi register imputed adl3
 
mi impute mlogit adl3 maxgrip i.country wave gender age_int age2 hhsize, replace rseed(2232) noisily

  midiagplots adl3
  midiagplots adl3, plottype(hist) separate
  mi xeq 0 1 10 20: summarize adl3
mi xeq 0 1 10 20: tab adl3