search

IALSA / ialsa-2016-groningen

Maelstrom Harmonization Workshop. Assessing the impact of different harmonization procedures on the analysis results from several real datasets.
GNU General Public License v2.0
1 stars 0 forks source link

Unrealistic values of estimates #20

Open andkov opened 8 years ago

andkov commented 8 years ago

@GracielaMuniz , @smhofer. The models that include interactions (AA and BB) report some unrealistic values of the estimates, for example for LBLS

educ3_f(HS):marital_f(Single)           4688189.56(0,1.40951620829273e+127) 912217.16(0,1.09476392250612e+137)
educ3_f(More than HS):marital_f(Single)         2252828.79(2.98254691622945e+26,5.28114380251839e+134)  244678.49(0,4.73769924409671e+132)

These estimation anomalies have been linked to the sparseness of data in the overlap of predictors: 

> d <- ds2 %>% dplyr::filter(study_name == "alsa")
> table(d$marital_f, d$educ3_f)

                  (HS) (Less than HS) (More than HS)
  (Single)          26             14             34
  (Married/Cohab)  546            200            606
  (Widowed)        222            116            240
  (Sep/Divorced)    22              6             21

> d <- ds2 %>% dplyr::filter(study_name == "lbsl")
> table(d$marital_f, d$educ3_f)

                  (HS) (Less than HS) (More than HS)
  (Single)           4              3             14
  (Married/Cohab)   97             27            188
  (Widowed)         39             32             45
  (Sep/Divorced)    15              4             55

> d <- ds2 %>% dplyr::filter(study_name == "satsa")
> table(d$marital_f, d$educ3_f)

                  (HS) (Less than HS) (More than HS)
  (Single)          10            113              9
  (Married/Cohab)   68            758             65
  (Widowed)         28            179             15
  (Sep/Divorced)     6             88             13

> d <- ds2 %>% dplyr::filter(study_name == "share")
> table(d$marital_f, d$educ3_f)

                  (HS) (Less than HS) (More than HS)
  (Single)          13             17             20
  (Married/Cohab)  699            737            586
  (Widowed)        114            156             54
  (Sep/Divorced)    57             48             54

> d <- ds2 %>% dplyr::filter(study_name == "tilda")
> table(d$marital_f, d$educ3_f)

                  (HS) (Less than HS) (More than HS)
  (Single)         153            327             40
  (Married/Cohab) 1498           2445            259
  (Widowed)        177            572             26
  (Sep/Divorced)   125            193             30