Thursday, October 23, 2008
Oct 28th HW
Consider the associations between children’s socio-economic status (SES) and their measured cognitive skills. Using the quintile version of SES (WKSESQ5), examine SES disparities in children’s standardized literacy (C1RSCALE) and math scores (C1MSCALE) at the beginning of kindergarten and 3 other continuous variables. Your table should display the post hoc results (use the middle quintile as the comparison group - see pg 16 of the summer learning paper).
Subscribe to:
Post Comments (Atom)
11 comments:
Hi Megan,
I've emailed Doug on this but no response yet and I need to work on hw tomorrow - I'd very much like to use the subcommand he mentioned that is needed to create a balanced categorical SES variable for the homework. If you could help? Many thanks,
Karen
Hi Megan,
Doug said the table should contain stars for significance relating to the middle SES level, which is fine. However, in writing about our results, is it ok to use a comparison group instead that is at one end of the range (lowest SES or highest SES), or do we need to be consistent with table? If I can use one end or the other it certainly simplifies expression.
Many thanks.
Karen
Hi Karen -
Sorry for the delay in replying - I'm out of town and access to email is limited. Here's my version of the syntax to create the new variable - it's a bit messy (and maybe Doug has something more streamlined, but it works):
***re-zscore wksesl .
DESCRIPTIVES VARIABLES=wksesl
/SAVE
/STATISTICS=MEAN STDDEV MIN MAX SEMEAN .
EXECUTE .
****check to make sure it worked .
DESCRIPTIVES VARIABLES=Zwksesl
/STATISTICS=MEAN STDDEV MIN MAX SEMEAN .
EXECUTE .
*********find the cut points for the new categorical .
FREQUENCIES
VARIABLES=Zwksesl
/NTILES= 5
/STATISTICS=STDDEV MINIMUM MAXIMUM MEAN
/ORDER= ANALYSIS .
******recode into a categorical based on the output from the syntax above .
RECODE
Zwksesl
(MISSING=SYSMIS) (Lowest thru -0.7722=1) (-0.7723 thru -0.3725=2) (-0.3726 thru 0.0804 =3) (0.0805 thru 0.8398=4) (0.8399 thru highest = 5) INTO Zses_cat .
VARIABLE LABELS Zses_cat 'Z-scored SES 5-categorical' .
VALUE LABELS Zses_cat 1 'low SES' 2 'med low SES' 3 'med SES' 4 'med high SES' 5 'high SES' .
EXECUTE .
********check to make sure it worked .
FREQUENCIES
VARIABLES=Zses_cat
/ORDER= ANALYSIS .
Karen - re:question 2, it is fine to talk about different comparison groups besides the middle one - whatever is interesting...
Megan
Hi Megan,
I am writing our analytical approach and I am wondering if this is an accurate statement:
"In terms of substantive significance we typically looked at the effect size of the outcome variable between the first and fifth and second and fifth quintiles"
Thanks,
Courtney
Hi Megan,
I am trying to put together our table using post hoc results, and my notes say to use the actual mean, not effect size. The mean differences in the post hoc tables are effect size, right? Then are we supposed to use the means listed in the descriptives tables?
Thanks!
Chrissy
Hello Megan,
A logic question. What, if anything, are we permitted to infer where there is no relationship between variables? Can we speculate in the conclusions about the cause of, e.g., a lack of relationship for a characteristic b/t the second and third quints?
Courtney -
I'm not sure what you mean by "between the first and fifth and second and fifth quintiles" - you should be looking at statistically significant differences in general, not necessarily limiting yourself to those two (?) comparisons.
I'm also not sure you need to go into that level of detail about substantive (vs. statistical) significance in this analytical approach section; the substantive significance really comes into play in your results section - how you write it up - I mean, there's no statistical "test" for substantive significance.
Chrissy -
Yes, use the means for each variable, not the mean differences, in your table.
Adam -
I think it's fine and potentially interesting to speculate about the reasons for the lack of a relationship, but I would say to leave it until your conclusions section.
This question is about the hw due Nov 4th. I ran the analysis both ways: using your "anova variables" syntax and using the menu to get the "unianova" procedure within GLM. (I figured I'd try the GLM since it was discussed in the book.) I got the same findings either way, but the F statistics and p values were not the same for both analyses. Is this normal? Using the anova variable procedure, I got an F=.407, p=.524 on the main effect of child care. Using the unianova procedure, I got an F=.052, p=.819 for the child care variable. Can you explain this?
Post a Comment