Re: test for trend

On 2 Feb 2006 10:21:22 -0800, raoulreulen@xxxxxxxxxxx wrote:

yes I did, sorry for my lack of knowledge regarding this. I have read
that you should use a dummy that is coded as -3 -1 +1 +3 in the
analysis (if you have 4 categories) which would give you a pvalue for
trend. Is this correct? Do you know any books that discuss this in more
detail? Thanks very much.

How big is your N? and is it equally apportioned to groups?

If your age-group Ns are not equal, then the usual formula
for orthogonal polynomials won't give orthogonal contrasts.
- but you can check the progressive contribution by putting
them in one-by-one. That is, you can use *any* version of
a linear trend, quadratic, cubic... if you evaluate them in sequence.

Using ANOVA, I've sometimes taken the SS of age groups and
pointed out, after the analysis, that the "linear component"
accounts for 95% (or whatever), of the total SS. That gives
an effect-size measure that might be indicative, even when
the sample N and total effect is large enough that the "non-linear"
component also tests as significant. For maximum likelihood
Logistic Regression, the chisquared contribution can be
considered in a similar fashion.

Ray's answer uses three dummy variables, and tells if
there is an increase with each increase in age. Is that
important, and are the Ns large enough to support a good
answer for asking the question that strictly?

If you just want a test for "generally it's a linear trend" across
the available Ns in all the groups, you can use (1,2,3,4) or
any linear coding.

--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.

Relevant Pages

  • RE: Trendlines to ignore empty cells
    ... That function could then be substituted for the TREND ... If the data covers only five points, the linear trend will return ... >> chart is to be placed “Test”. ... >> Assume the letter “a” starts in cell A2. ...
    (microsoft.public.excel.charting)
  • Re: test for trend
    ... > How do I perform a test for linear trend of a variable that is coded as ... > let's say I have the variable age which is categorized into 4 groups. ... Use echelon coding: grp age1 age2 age3 ...
    (sci.stat.consult)
  • Re: Edward Prescott, Idiot
    ... REAL GDP GROWTH IS MEAN REVERTING. ... Most time series methods used ... shows a linear trend. ... trend stationary with a break. ...
    (sci.econ)
  • test for trend
    ... How do I perform a test for linear trend of a variable that is coded as ... let's say I have the variable age which is categorized into 4 groups. ... use the lowest category as the reference catogory. ...
    (sci.stat.consult)
Loading