Reaction time outliers in e-prime

[ashraf]

hi group

what the effective way to remove reaction time outliers by e-prime
and ,should i analysis the removable values when i analysis the
accuracy
ashraf

[David McFarlane]

ashraf,

Offhand, I don't know if this is a question about how to do something
in E-Prime, or a more general question about how to handle outliers,
which would be better suited for a class on data analysis. Also, is
there any reason to have E-Prime handle outliers at run time rather
than to gather all the raw data first and then deal with outliers
during analysis? And you would probably use some statistics analysis
package to handle the outliers, not E-Prime. But first you have to
have a good grounding in statistical data analysis before you start
throwing out data, and I would think that that topic goes way beyond
the purpose this E-Prime Group.

Regards,
-- David McFarlane, Professional Faultfinder
"For a successful technology, reality must take precedence over
public relations, for nature cannot be fooled." (Richard Feynman,
Nobel prize-winning physicist)

[ashraf]

David McFarlane,
thank you for your reply ,there is a filter in e-prime could help me
to handle outliers in reaction time before analysis,my question about
that,also about analysis the accuracy in e-merge. please help
me ,thank you very much

[liwenna]

Hi Ashraf,

I think that many users don't really use e-merge for their data-
analysis (I don't at least and I know noone who does). So I really
couldn't tell you how to do that although I guess it would be similar
to the way you would do it when for instance using excel or spss.

What I myself do: I merge the edat files and export them to a .txt
file which I then import in SPSS. The next step is to know how you
want to define outliers: do you want to define outliers per subject,
per condition, per group etc etc. And do you want to define an outlier
as more than 2 sd's from the mean or 3 sd's from the mean? In many
tasks there's also a 'raw' cutoff of scores used based on the idea
that for instance (and depending on your task) any reactiontime under
say 200 ms or over 2000 ms are probably due to the subject not paying
attention to the task enough (i.e. reasonably too fast or too slow).
Once you have your definition you calculate the means and sd's over
the collection of reactiontimes that you want to remove outliers from
(for instance mean per group or mean per subject or mean per subject
AND condition etc). and then calculate the lower and upper bound (by
adding and subtracting 2 or 3 or depending on your definition sd's
from the mean) and then filter the reactiontimes under and over these
bounds out for that collection of reactiontimes. Alternatively you
could use median scores instead of mean scores for which you would not
need to calculate outliers... but... this might be harder to get
published, yet for a first glance at your data is a quicker way.

I agree with David that this is definitaly beyond the scope of this
google group which has more to do with using e-prime the program as
with the analysis of data collected with e-prime. In order to come to
a definition of your outliers you ought to consult with existing
literature on your task, with your statistics books and your common-
sense (and your supervisors perhaps? If you have one that is).

Good luck on it though :)

liw

[David McFarlane]

ashraf,

Mich already kindly gave some specific instructions on how to use the
filtering feature of E-DataAid. For more information, please work
through the the E-DataAid tutorial in the Getting Started Guide, the
Data Handling chapter in the User's Guide, and the E-DataAid chapter
of the Reference Guide that came with E-Prime. While you are at it
you should also study the E-Merge chapters in those Guides. It is
well worth the effort, in my experience these tools are vastly
underutilized even though they provide enough benefit that it is
almost worth using E-Prime for these data analysis tools alone. They
still will not do statistical hypothesis testing or more complicated
data manipulation for you, but with a little effort these tools can
*greatly* simplify the early stages of data analysis.

-- David McFarlane, Professional Faultfinder

[David McFarlane]

>but with a little effort these tools can *greatly* simplify the
>early stages of data analysis.

Just to belabor this a bit, I refer specifically to the Analyze and
Batch Analysis features of E-DataAid, which get overlooked all too
much (took me a decade to discover these myself!).

[ashraf]

hi groub,
thank you very much for your interest,