Skip to content

intercept value in lme

5 messages · Doran, Harold, Chuck Cleland, Andrew Robinson +1 more

#
As Andrew noted, you need to provide more information. But, what I see
is that your model assumes X is continuous but you say it is bounded,
-25 < X < 0
#
It is boundend, you're right. In fact it is -25<=X<=0

These are cross-national survey data (I was investigated 7 countries in 
each country there was 900-1700 cases).
In fact, there was two level 2 variables, so:

m1<-lme(X~Y,~1|group,data=data,na.action=na.exclude,method="ML")
m2<-lme(X~Y+Z1+Z2,~1|group,data=data,na.action=na.exclude,method="ML")

X is a life satisfaction factor combined from 2 other variables for each 
case separately, of course.
Y  - income per capita in household
Z1 - unemployment rate in a country.
Z2 - life expectancy in a country
group - country

I attach a similar model where after adding Lev2 predictors intercept 
value is even 22!

I'm sure there is my mistake somwhere but... what is wrong?



Linear mixed-effects model fit by maximum likelihood
  Data: data
        AIC      BIC    logLik
   31140.77 31167.54 -15566.39

Random effects:
  Formula: ~1 | country
         (Intercept) Residual
StdDev:   0.8698037 3.300206

Fixed effects: X ~ Y
                 Value Std.Error   DF    t-value p-value
(Intercept) -4.397051 0.3345368 5944 -13.143698       0
Y           -0.000438 0.0000521 5944  -8.399448       0
  Correlation:
         (Intr)
Y       -0.13

Standardized Within-Group Residuals:
        Min         Q1        Med         Q3        Max
-6.3855881 -0.5223116  0.2948941  0.6250717  2.6020180

Number of Observations: 5952
Number of Groups: 7


and for the second model:

Linear mixed-effects model fit by maximum likelihood
  Data: data
        AIC      BIC    logLik
   31133.08 31173.23 -15560.54

Random effects:
  Formula: ~1 | country
         (Intercept) Residual
StdDev:   0.3631184 3.300201

Fixed effects: X ~ Y + Z1 + Z2
                 Value Std.Error   DF   t-value p-value
(Intercept) 22.188828  4.912214 5944  4.517073  0.0000
Y           -0.000440  0.000052 5944 -8.456196  0.0000
Z1          -0.095532  0.037520    4 -2.546161  0.0636
Z2          -0.333549  0.062031    4 -5.377127  0.0058
  Correlation:
         (Intr) FAMPEC UNEMP
Y        0.168
Z1      -0.429  0.080
Z2      -0.997 -0.188  0.366

Standardized Within-Group Residuals:
        Min         Q1        Med         Q3        Max
-6.3778888 -0.5291287  0.2963226  0.6260023  2.6226880

Number of Observations: 5952
Number of Groups: 7
Doran, Harold wrote:
#
victor wrote:
Victor:
  What happens if you center Y, Z1, and Z2 so that 0 corresponds to the
mean for each?  As it is, zero is a very unusual value for each of these
variables.  Do you really want to estimate the value of X when income =
0, unemployment = 0, and life expectancy = 0?  If I understand
correctly, I think that's why the intercept value looks unusual to you.

  
    
#
Hello Victor,

I'm afraid that this still isn't what we're looking for, in terms of
reproducible code, but we can guess.  What is the range of the 
Z1 and Z2 variables?  What is the range of the model predictions? 
If the Z1 and Z2 variables are large and positive then they will be
compensating.

Cheers

Andrew
On Wed, Dec 06, 2006 at 06:06:55PM +0100, victor wrote:

  
    
#
Thanks to all of you!
Yes, you're right - I didn't take into consideration the ranges of 
predicors which are quite large. I think the matter over and realize 
that my assumption that something have to be wrong doesn't have in fact 
any reason except "strange" look of the value.
Centering helped (as suggested by Chuck) especially in interpretation 
and helped me to understand what is really going on in the model.

Thank you once again - these are my first experiences with R as like as 
with multilevel models, so... thank you for your patience!

Best regards,

victor
Chuck Cleland wrote: