Date: Tue, 24 Oct 2006 17:21:38 -0400
Reply-To: "Howard Schreier <hs AT dc-sug DOT org>" <nospam@HOWLES.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: "Howard Schreier <hs AT dc-sug DOT org>" <nospam@HOWLES.COM>
Subject: Re: How can I do output like this....
On Mon, 23 Oct 2006 01:19:56 -0700, Fei <joey_qf@163.COM> wrote:
>Thank you so much David!
>Originally, my data was in Excel and each spreadsheet contains one
>company's data, then I import each spreadsheet into SAS! 100 Companies
>were imported into 100 SAS sheet! Do I have a way to import all once -
>they are named as code of companies, so different names?
>Since they are in separated SAS sheets, I run regression for each one
>separately (I just copy and paste code and change to new company name)
>and then run regression for different periods - I use IF statement to
>choose appropriate period, e.g if date>20000101, that is I want to data
>before 2000-01-01. All my regressions are separated. I just a beginner,
>so do not know too complicated stuff.....Sorry to trouble you! I know
>my way is quite stupid...
It's a good idea to get your data arranged and organized before plunging
into the analysis or other purpose. Ultimately, it saves time.
Details are important here. Do you have one XLS workbook with 100 tabs? 100
workbooks with one tab in each? All files in the same directory? What SAS
version? Is SAS Access for PC File Formats licensed and installed?
>My model is Y=a+bx1+cx2 - all sheets have the same dependent and
>independent variables, say all regressions use capitalisation as
>dependent variable, and return on equity and number of employees as
>independent variables (they are just examples, so may not make sense to
>Now I would like to put coefficients of return on equity (ROE) and
>number of employees across different companies and period in a table
>(output as a SAS dataset) - if it is possible I also want P-value and
>F-value in the table and table likes:
> coef of ROE P-value_ROE coef of No
>Emp P-Value_Emp F-value
>coef of ROE P-value_ROE coef of No Emp P-Value_Emp F-value are
>titles - the message cannot fit them in one line.
>Thank you so much!
>David L Cassell wrote:
>> joey_qf@163.COM wrote:
>> >Thank you David!
>> >You are right I have to think through all again - I have 4 specific
>> >groups and just want pair up these groups in whose coefficient is
>> >closest to each four! The table is too huge!
>> >At this point, I think picking up these comparable groups to the four
>> >specific groups manually may even quicker, since my dataset up is
>> >different with code Howard suggested - I put each group in a separated
>> >sheet! Subset cannot apply to mine!
>> >Thank you so much for all eth helps!
>> First, I think that you need to investigate how to estimate your
>> 'slope' if you have time series data that might not meet all the
>> assumptions of basic regression. Think about it. Are your data
>> increasing over time because you have a positive slope, or are
>> they increasing because they are positively autocorrelated, or
>> is the underlying model even more complex than that?
>> Second, I think that you need to investigate whether slope
>> alone is a reasonable way to decide how to distribute your
>> values into the four classes.
>> Third, I think that you need to investigate statistical ways to
>> cluster your data into these classes. I said 'cluster', because
>> this might be a cluster analysis problem. Doing this by eye
>> may be adequate for a single variable where your classes have
>> lots of separation, but even there you will run into trouble
>> as soon as you have overlapping classes, classes with unequal
>> variability, classes with unequal separation, ...
>> So I think your problem is even bigger than I guessed from
>> your first post. Here are some of the questions you need to answer:
>> Why do you want to do this?
>> What is it for?
>> Where do you go once you get your 4 groups?
>> What should you do about time series data?
>> What should you do about outliers?
>> What should you do about leverage points?
>> What should you do about other regression assumptions?
>> How should you estimate those slopes?
>> What do the slopes actually *mean* ?
>> How should you 'cluster' your groups?
>> David L. Cassell
>> mathematical statistician
>> Design Pathways
>> 3115 NW Norwood Pl.
>> Corvallis OR 97330
>> Try the next generation of search with Windows Live Search today!