Anonymous posted on Wednesday, June 02, 2004 - 5:09 pm
I'm using v2.14 and I am estimating a growth model where time is based on the day of observation. That calculation works fine. However, I would like to add a second time variable (the value of time squared) along with the first. In SAS I would do this by including the command "random Intercept time time2." Is there anyway to do this in Mplus.
I think you are using the AT command. If so, to add a quadratic component, say
i s1 s2 | y1 y2 y3 y4 y4 AT t1 t2 t3 t4;
This is described on page 40 of the Addendum to the Mplus User's Guide. i is the intercept growth factor, s1 is the linear growth factor, and s2 is the quadratic growth factor.
Anonymous posted on Friday, November 11, 2005 - 6:17 am
Dear Drs Muthen,
I have data on a physical activity measure (continuous) on about 100 people measured annually over 3 years, with the aim to assess changes over time and predictors of it. However, the ages at first measurement vary enormously (age range 40-80 years). I was hoping to use the SEM-style wide format in Mplus to get all the fit indices etc, but there are many distinct ages. Is there a way to handle this age variation in wide format?
I have fit a multilevel random slopes model, but the question of time centering and time metric arose, as using age (age-60) as the time metric assumes linearity over the whole age range (which I don't really want to assume), and produces a correlation between intercept and slope of about -0.80 at age 60, and close to one for more complicated models. I have thought about centering time at the first measurement for each person to alleviate this, but is there another option in Mplus to reduce this numerical problem? Also, do you know of any references for these sort of decisions with hugely variable initial ages and only a few relatively closely-spaced times of measurement?
bmuthen posted on Sunday, November 13, 2005 - 4:38 pm
If you use the wide, multivariate approach I think you would have to use the multiple-cohort, multiple-group approach where each age category (say 40-50 year olds being one category) corresponds to a group. In this approach you could also test the key assumption that the same growth model holds across all these quite different ages. But you may not have enough individuals within each group for this approach. Otherwise, you would have to use a 2-level, long data format, approach where you read in age as a variable in line with conventional multilevel growth modeling. I don't know of any pertinent references here.
Anonymous posted on Wednesday, October 11, 2006 - 9:49 pm
I would like to use number of months (or years) that a caregiver has been providing care as a time metric. Participants respond to this question at baseline and then respond to additional surveys 3 and 12 months later. The range in months for this variable is 0-288. If I convert this to years and winsorize these data so that each person has a range of 0-10 years providing care, could I then use categories that correspond to each group (e.g., 0-3 yrs. = 0) as a metric? I would like to use a 2-level growth model. Otherwise, could you suggest an alternative solution? Thanks!
Your proposal sounds reasonable - using an alternative time metric like you suggest is sometimes more meaningful.
Anonymous posted on Thursday, October 12, 2006 - 3:30 pm
Thank-you. As I prepare these data for use in MPlus, how should I assign time scores for each person at each of the 3 measurement occasions? Unless participants move into another group at 12-months, each person's time score at baseline would remain stable (i.e. 0, 1, or 2) across the occasions.