This is a general observation:
When a client walks into my office, hands me a disk-on-key of data with the
instructions "please run three T-test and two Chi-square's" my first
inclination is to show him to the door.
After I count, slowly, till ten; I attempt to explain that if he would take
time to fully familiarize me with the research question, I would be able to
more fully assist. Usually it turns out that the preliminary analysis
requested is sub-optimal.
When I am on very good terms with said client, I suggest that seeking
statistical aid at the very onset of the research - when the study question
is first formulated and before any data have been collected; may serve to
advert future problems.
It never ceases to amaze me how little time of a research program is
actually spent on question formulation, yet countless examples exist of
serious people who spent large amounts of time and money answering the wrong
question.
Tzippy
----- Original Message -----
From: "Ali A. Bromideh" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Saturday, July 30, 2005 2:35 PM
Subject: Query: How to analysis a panel data.
> Dear Alstat!
> I've been asked to share this question with the AllStat group. Please
> reply if you have an idea in this regard.
>
> Introduction and problem statement:
> A researcher was interested in explanation of significant relationship
> between four variables named Y (dependent) and X1, X2 and X3
(independents).
> For 30 SMEs (Small/Medium Size Enterprise), he has collected data from
> 1999-2004. Now he wanted to analyze the data at hand. He has been asked to
> use these techniques:
> 1- Put all data in a data set and regardless of time, run a "correlation
> analysis." That means, for each company we will have 5 observation and
total
> sample size will be 5*30=150. In other words, he will calculate the
> correlation coefficient between the variables.
> 2- Calculate the average for the five years for each variable and run a
> correlation analysis, based on these averages. In other words, first he
> reduces the data by taking mean value and then again a correlation
analysis
> (in this case N=30).
> 3- Use a dummy variable for time and then use advanced statistical
> methodology, like Longititudinal or Panel data.
>
> Question:
> Would you explain a bit more about the pros and cons of the proceedings
> methods, please?
>
> Kind regards,
> ALI
>
|