Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

# Re: st: Intraclass Correlation for an Independent Variable?

 From Steve Samuels To statalist@hsphsun2.harvard.edu Subject Re: st: Intraclass Correlation for an Independent Variable? Date Thu, 12 Aug 2010 13:26:10 -0400

```Well, I will say this:  The data show that plant means in the worker
measurement X differ no more than would be expected if workers had
(say) been randomly assigned to plants. How the differences in plant X
means arose is immaterial to whether they affect plant outcome. I
think that is a response  that you can give the reviewer.
.
To my mind,  including a summary measure such as the mean of X is
perfectly legitimate. Trying to find the most predictive summary X is
okay, as long as you are trying to control for the possible effects of
X, not test hypotheses about it. Otherwise you have to quote the
p-values and coefficients for all the summaries you consider and make
some statement about the multiple test problem.

Steve

On Thu, Aug 12, 2010 at 12:43 PM, Steve Samuels <sjsamuels@gmail.com> wrote:
> Lloyd, Whether, and how, to include the worker measurements as
> predictors of plant outcomes are substantive questions that I don't
>
> Steve
>
> On Thu, Aug 12, 2010 at 12:22 PM, Lloyd Dumont <lloyddumont@yahoo.com> wrote:
>> Thank you for your advice on this, Steve.  From a technical/statistical point-of-view, it was very helpful.  I ran -loneway- on the dataset of individual workers.  The F-stat was about 1.24, generating a p-value far greater than I would have wanted in order to please the reviewer--about .24.
>>
>> Given that, is there anything I can do to make the case for these measures?  Is there a way to "correct for" the heterskedasticity of the estimated means for each plant before collapsing them into a dataset of plant-months?  Off the top of my head, the first thing I thought of was to -collapse- the data using medians as opposed to means.  That way, I could at least be assured that outliars are not influencing each plant's score.  Interestingly, when I did that, the results in the regression of plant-months became a little stronger.
>>
>> Thank you again.  Lloyd
>>
>> --- On Wed, 8/11/10, Steve Samuels <sjsamuels@gmail.com> wrote:
>>
>>> From: Steve Samuels <sjsamuels@gmail.com>
>>> Subject: Re: st: Intraclass Correlation for an Independent Variable?
>>> To: statalist@hsphsun2.harvard.edu
>>> Date: Wednesday, August 11, 2010, 5:47 PM
>>> I didn't read the second part of your
>>> message closely enough.  ICCs
>>> can be computed for any variable measured in groups,and the
>>> reviewer's
>>> terminology is impeccable.  -loneway- should do what
>>> you want. But
>>> "search intraclass, all" will find some extensions.
>>>
>>> Steve
>>>
>>> On Wed, Aug 11, 2010 at 5:05 PM, Steve Samuels <sjsamuels@gmail.com>
>>> wrote:
>>> > The reviewer could well want to know if the plant
>>> differences in the X
>>> > means are real.  So, yes, he or she could really mean
>>> "X".  This is
>>> > the kind of question you'd test with a one-way ANOVA,
>>> or KW. You will
>>> > need the data from the individual worker X
>>> measurements, or the
>>> > plant-specific sample sizes and  variances.
>>> >
>>> > Steve
>>> > On Wed, Aug 11, 2010 at 4:36 PM, Lloyd Dumont <lloyddumont@yahoo.com>
>>> wrote:
>>> >> Hello.  I have gotten a confusing note from a
>>> reviewer, confusing in part because I think he/she is using
>>> the wrong terminology.
>>> >>
>>> >>
>>> >> In a nutshell…
>>> >>
>>> >> I am running a cross-sectional time series model
>>> (on a dataset of plant-months) of a continuous variable Y
>>> measured at the plant level where the focal, continuous
>>> independent variable is X.  X is also at the plant-level,
>>> but is time-constant.  X is formed by taking the mean level
>>> of X from all of the individual workers in each plant.
>>> >>
>>> >> The reviewer suggests that I include some measure
>>> of within versus between plant variance in X.  This would
>>> bolster the case that there is a genuine difference across
>>> plants rather than average differences simply reflecting
>>> noise.  He/she then goes on to suggest I do this by
>>> reporting intraclass correlations (ICCs).
>>> >>
>>> >> As far as I understand it, ICCs (or rho) describe
>>> the variance in Y, not X.  If I am right, then the reviewer
>>> did not mean ICC.
>>> >>
>>> >> Rather, he/she wants me to go back to the dataset
>>> of individuals (not plant-months) and wants me to run some
>>> simple analysis to show that there is clustering of values
>>> for X by plant.
>>> >>
>>> >> Am I correctly understanding what the reviewer is
>>> asking for?  And, if so, what is the simplest way to
>>> demonstrate the sort of dependence he/she is hoping to see?
>>> >>
>>> >> Thank you for your help.  Lloyd
>>> >>
>>> >>
>>> >>
>>> >>
>>> >>
>>> >>
>>> >> *
>>> >> *   For searches and help try:
>>> >> *   http://www.stata.com/help.cgi?search
>>> >> *   http://www.stata.com/support/statalist/faq
>>> >> *   http://www.ats.ucla.edu/stat/stata/
>>> >>
>>> >
>>> >
>>> >
>>> > --
>>> > Steven Samuels
>>> > sjsamuels@gmail.com
>>> > 18 Cantine's Island
>>> > Saugerties NY 12477
>>> > USA
>>> > Voice: 845-246-0774
>>> > Fax:    206-202-4783
>>> >
>>>
>>>
>>>
>>> --
>>> Steven Samuels
>>> sjsamuels@gmail.com
>>> 18 Cantine's Island
>>> Saugerties NY 12477
>>> USA
>>> Voice: 845-246-0774
>>> Fax:    206-202-4783
>>>
>>> *
>>> *   For searches and help try:
>>> *   http://www.stata.com/help.cgi?search
>>> *   http://www.stata.com/support/statalist/faq
>>> *   http://www.ats.ucla.edu/stat/stata/
>>>
>>
>>
>>
>>
>>
>> *
>> *   For searches and help try:
>> *   http://www.stata.com/help.cgi?search
>> *   http://www.stata.com/support/statalist/faq
>> *   http://www.ats.ucla.edu/stat/stata/
>>
>
>
>
> --
> Steven Samuels
> sjsamuels@gmail.com
> 18 Cantine's Island
> Saugerties NY 12477
> USA
> Voice: 845-246-0774
> Fax:    206-202-4783
>

--
Steven Samuels
sjsamuels@gmail.com
18 Cantine's Island
Saugerties NY 12477
USA
Voice: 845-246-0774
Fax:    206-202-4783

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```