If this is your first visit, be sure to check out the FAQ by clicking the link above. You may have to register before you can post: click the register link above to proceed. To start viewing messages, select the forum that you want to visit from the selection below. |
|
|
Thread Tools | Display Modes |
#1
|
|||
|
|||
Regression difficulty
When doing a regression analysis I am constrained with the
use of only 16 explanitory variables. Does anyone know anyway around this? Thanks, Zak |
#2
|
|||
|
|||
Regression difficulty
Use a dedicated stats package, such as S-Plus (Windows-only) or R
(open-source, free equivalent of S, available also in compiled form for many platforms. There are certainly other packages, but those are my favorites, in that they are not exactly dedicated to stats, but rather object-oriented environments that are stat-focused, so to speak. For R, which I highly recommend, see SourceForge. In the future, consider including the version of Excel you have in mind, as the stat functionality has changed quite a lot in the last couple of versions. HTH, Dave Braden MVP - Excel In article , "Zak" wrote: When doing a regression analysis I am constrained with the use of only 16 explanitory variables. Does anyone know anyway around this? Thanks, Zak -- (ROT13) |
#3
|
|||
|
|||
Regression difficulty
In message , Zak
writes When doing a regression analysis I am constrained with the use of only 16 explanitory variables. Does anyone know anyway around this? Excel's original built in regression functions only just work reliably for the case of 3 explanatory variables for ill conditioned problems. You are crazy if you use it for anything where getting the right answer actually matters. The MS mantra on statistics is that "business users do not care". Look for previous threads comparing the performance of the spreadsheet function linear regression against the chart polynomial fit for more details. It is claimed that they have finally improved the situation in Office XP but I haven't got around to testing it. Perhaps others have? Use a decent statistical package instead and even then always check that the notional fit it computes really is a true least squares solution! Regards, -- Martin Brown |
#4
|
|||
|
|||
Regression difficulty
Martin,
I agree that the MS mantra you cite has been true since Excel first appeared (on a Mac, at that); that mantra is being redefined, so to speak. Excel 11 has marked improvement in LINEST, TREND, SLOPE, and such, as well as the VAR-related functions. Some of the probability functions (such as those related to the Normal) are much better, indeed IMHO acceptable for serious work. Several of us have put these functions through the wringer. As you know, I am hardly an MS toady, yet can honestly write that, to their credit, MS has really worked hard to address some of the stat shortcomings that you, others and I have been long pointing out. To get a sense of where MS's mindset has come, check out http://support.microsoft.com/?kbid=828888 If you concede that the company is loathe to admit problems with Office, then you must admit that this KB article (sheesh, site) says a lot about what's being done "behind the scenes" As for your suggestion re using a decent stat package, I couldn't agree more. For critical stuff, I always use at least two packages. R is really coming along at a fast clip, and is free. Regards, Dave Braden MVP - Excel In article , Martin Brown wrote: In message , Zak writes When doing a regression analysis I am constrained with the use of only 16 explanitory variables. Does anyone know anyway around this? Excel's original built in regression functions only just work reliably for the case of 3 explanatory variables for ill conditioned problems. You are crazy if you use it for anything where getting the right answer actually matters. The MS mantra on statistics is that "business users do not care". Look for previous threads comparing the performance of the spreadsheet function linear regression against the chart polynomial fit for more details. It is claimed that they have finally improved the situation in Office XP but I haven't got around to testing it. Perhaps others have? Use a decent statistical package instead and even then always check that the notional fit it computes really is a true least squares solution! Regards, -- (ROT13) |
Thread Tools | |
Display Modes | |
|
|