SPSS vs Statistca
Locked
 
|
I wonder if anyone on the list can provide an opinion on the differences of SPSS and Statistca. I have a learner who wants to use Statistca stating that is it a much better option that SPSS because it is more accurate. I can't seem to locate any information to support this view. Any assistance would be greatly appreciated.
Dana Dana Barber Gonzales, Ph.D Assistant Professor Department of Dietetics and Nutrition University of Arkansas for Medical Sciences 4301 West Markham St. Slot # 627 Little Rock, AR 72205 Phone # 501-686-6166 Fax # 501-686-5716 ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Re: SPSS vs Statistca
|
|
Gonzales, Dana L wrote:
> I wonder if anyone on the list can provide an opinion on the > differences of SPSS and Statistca. I have a learner who wants to use > Statistca stating that is it a much better option that SPSS because > it is more accurate. I can't seem to locate any information to > support this view. Any assistance would be greatly appreciated. I'm not sure you'll get unbiased opinions from people who subscribe to this list. But perhaps some of the opinions offered might still be helpful. Here are my thoughts: More accurate is a surprising choice of words. If this person means numerical accuracy, that's pretty much a non-issue for most problems as all professional computer programs have ditched single precision and avoided the lousy algorithms. It's best not to mention Microsoft Excel at this point. That doesn't mean that you can't trip up one of these programs with a tricky data set, but in general, accuracy is not a serious concern. Several years ago, Statistica had a very aggressive advertising campaign that cast aspersions on many of its competitors. I think that there was also some pushback from people like Leland Wilkinson that held the claims to be unfair and unsupported. I wonder if this person has some residual memory of this campaign. If "accuracy" means something other than numerical accuracy, then you need to define what this really means. I could make some guesses, but that's dangerous. I would discourage a comparison of two statistical packages based on accuracy. The criteria that make more sense are: (1) ease of use, and (2) availability of advanced procedures. I have no experience with Statistica, but I'd be surprised based on what I have read about the program, that it would be considered vastly superior in either category. -- Steve Simon, Standard disclaimer. Looking for new friends/connections on Facebook and LinkedIn www.facebook.com/pmean, www.linkedin.com/in/pmean or become a fan of my newsletter, The Monthly Mean http://www.facebook.com/group.php?gid=302778306676 ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Effects of the Dichotomous variables on a Dichotomous Outcome
|
|
Cleaner, Better, Faster - Experience the new Faster Yahoo! Mail today! |
Re: Effects of the Dichotomous variables on a Dichotomous Outcome
|
|
Eins Bernardo wrote:
> (Sorry for cross-posting) > > I want to know which of the variables X1, X2, X3, X4 and X5 are > significant predictors of the outcome variable Y where both Y and the > X's are dichotomous (coded 0 for NO and 1 for YES). The sample size > is n=40. The target variable Y has 32 for YES and 8 for the NO response. > > Can I use logistic regression? Any comments are welcome. > Hi Eins: If you are talking about 5 univariate logistic regression modeles (Y with X1, Y with X2...), go ahead, or use contingency tables (exactly the same). If you are planning to run a multivariate logistic regression model (Y with X1 and X2 and ... X5), the answer is NO. Not with that sample size (32 yes + 8 no). Marta -- For miscellaneous SPSS related statistical stuff, visit: http://gjyp.nl/marta/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Re: Effects of the Dichotomous variables on a Dichotomous Outcome
|
|
In reply to this post by E. Bernardo
What is your suggestion in terms of sample size and proportion of the yes
and no response? Is the sample size requirement follows 10 to 20 cases per independent variable? There was a comment in this list that the proportion of a yes response to the no response is not an issue. ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Re: Effects of the Dichotomous variables on a Dichotomous Outcome
|
|
In reply to this post by Marta Garcia-Granero
Yahoo! Mail Now Faster and Cleaner. Experience it today! |
Re: Effects of the Dichotomous variables on a Dichotomous Outcome
|
|
Hi,
Eins Bernardo wrote: > (Sorry for cross-posting) > > I want to know which of the variables X1, X2, X3, X4 and X5 are > significant predictors of the outcome variable Y where both Y and the > X's are dichotomous (coded 0 for NO and 1 for YES). The sample size > is n=40. The target variable Y has 32 for YES and 8 for the NO response. > > Can I use logistic regression? Any comments are welcome. You can use Logistic regression since the outcome is a yes-no response. However, do note something
Is it possible for you to collect more information for the No response? Do take note of the usual checks in the Logistic Regression output, Omnibus Tests, Nagelkerke R Square, Hosmer and Lemeshow Test, etc. If in need and even possible, please do collect more data on the No-Responses. Regards Dorraj Oet Date: Sun, 7 Mar 2010 21:59:33 +0800 From: [hidden email] Subject: Re: Effects of the Dichotomous variables on a Dichotomous Outcome To: [hidden email]
Yahoo! Mail Now Faster and Cleaner. Experience it today! Hotmail: Trusted email with Microsoft’s powerful SPAM protection. Sign up now. |
Re: Effects of the Dichotomous variables on a Dichotomous Outcome
|
|
In reply to this post by E. Bernardo
Eins Bernardo escribió:
> What is your suggestion in terms of sample size and proportion of the yes > and no response? Is the sample size requirement follows 10 to 20 cases > per independent variable? There was a comment in this list that the > proportion of a yes response to the no response is not an issue. > > > At least 10 to 20 "Yes", or "10 to 20" No, whatever is less frequent, per independent variable, therefore, you need 50 to 100 "yes" and 50 to 100 "no", since you are studying 5 independent predictors. Rules of thumb are, of course, only approximate, not absolute dogmas. HTH, Marta GG -- For miscellaneous SPSS related statistical stuff, visit: http://gjyp.nl/marta/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Re: Effects of the Dichotomous variables on a Dichotomous Outcome
|
|
In reply to this post by E. Bernardo
Eins Bernardo wrote:
> Marta, et al. > > I am using contingency table to analyze the data then try with Simple > Logistic Regression. Like Marta, prior to analysis, I have in mind > that the two analyses are the same. However, I noticed that they > produced different results. The logistic regression revelaed that > the X is not significantly associated with the Y. The Fisher exact > test and the contingency coefficient revealed that X is significantly > associated with Y. > > The data are as follows: > > X Y Count > yes yes 13 > yes No 0 > No No 8 > No Yes 19 > Did you use Wald test? (in logistic regression). It has low power. You should use LR test. Marta > ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Re: Effects of the Dichotomous variables on a Dichotomous Outcome
|
|
In reply to this post by E. Bernardo
Try the new FASTER Yahoo! Mail.. Experience it today! |
|
|
In reply to this post by Marta Garcia-Granero
Adding more friends is quick and easy. Import them over to Yahoo! Mail today! |
Re: Effects of the Dichotomous variables on a Dichotomous Outcome
|
|
In reply to this post by E. Bernardo
Eins Bernardo wrote:
> > I am using contingency table to analyze the data then try with Simple > Logistic Regression. Like Marta, prior to analysis, I have in mind > that the two analyses are the same. However, I noticed that they > produced different results. The logistic regression revelaed that > the X is not significantly associated with the Y. The Fisher exact > test and the contingency coefficient revealed that X is significantly > associated with Y. > > The data are as follows: > > X Y Count > yes yes 13 > yes No 0 > No No 8 > No Yes 19 > You CAN'T use logistic regression when you have a zero in one cell. You get an indeterminate OR (13*8/(19*0), and a SE(log(OR))=SQRT (1/13+1/8+1/19+1/0), that tends to infinity. That's why you get a non-significant Wald test, although you do get a very significant LR test (and you also get significant results when using contingency analysis tables). This example clearly shows that your sample size is definitely too low. You can get an approximate estimate of the OR by adding 0.5 to each cell frequency: DATA LIST LIST/ X Y (2 A3) Count (F8). BEGIN DATA yes yes 13 yes no 0 no no 8 no yes 19 END DATA. COMPUTE Count=Count+0.5. WEIGHT BY Count. LOGISTIC REGRESSION VARIABLES Y /METHOD = ENTER X /CONTRAST (X)=Indicator(1). Anyway, Wald's test power is still too low, and the result is non significant. -- For miscellaneous SPSS related statistical stuff, visit: http://gjyp.nl/marta/ ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
|
|
In reply to this post by Johnny Amora
If you install the R plugin for V16, the documentation installed with it and can be accessed from the Help menu. Although V17 and V18 have many api enhancements, V16 provides access to the active SPSS data and the ability to create SPSS output pivot tables from R. The R-based extension commands on Developer Central, however, require a later SPSS version. HTH, Jon Peck SPSS, an IBM Company [hidden email] 312-651-3435
Adding more friends is quick and easy. Import them over to Yahoo! Mail today! |
|
|
In reply to this post by Steve Simon, P.Mean Consulting
I work in a SPSS / Statistica office, IMHO Statistica is not the software of choice for any serious analyst. As the previous poster pointed out, precision is a moot point. I've used SPSS for over 15 years and while there is the occasional bug there has been no major issue regarding precision of results with any major software package, SPSS included, that I am aware of. While Statistica may be easy to use, if you define easy to use as point and click, the seemingly endless array of point-and-click menus becomes a liability after a point because the sheer number of options etc that need to be clicked becomes very confusing and makes it difficult to replicate a procedure or analysis. Which leads to the primary shortcoming with Statistica: there is no syntax option. Thus there is no way to build audit trails to document complex analyses. I've never worked in a shop that relied on point and click software because it is imperative to document / replicate work. Not to mention syntax greatly simplifies repetitive tasks and often enables a degree of customization not available through menus. Also, I have found that Statistica does not play nicely with other software, i.e., moving
data into or out of Statistica often results in scrambled variable names / labels / value labels / variable types etc.. Finally, I have found data management in Statistica (merging files, sub setting cases, etc.) to be unusually difficult relative to SPSS (or SAS or SQL). Statistica is probably OK for teaching statistics because you can eliminate the layer of confusion that can result from the simultaneous teaching syntax and statistics. But for professional work I would choose almost anything but Statistica. Obviously this is all based on my personal experiences, perhaps others have had better. From: "Steve Simon, P.Mean Consulting" <[hidden email]> To: [hidden email] Sent: Sat, March 6, 2010 10:49:24 PM Subject: Re: SPSS vs Statistca Gonzales, Dana L wrote: > I wonder if anyone on the list can provide an opinion on the > differences of SPSS and Statistca. I have a learner who wants to use > Statistca stating that is it a much better option that SPSS because > it is more accurate. I can't seem to locate any information to > support this view. Any assistance would be greatly appreciated. I'm not sure you'll get unbiased opinions from people who subscribe to this list. But perhaps some of the opinions offered might still be helpful. Here are my thoughts: More accurate is a surprising choice of words. If this person means numerical accuracy, that's pretty much a non-issue for most problems as all professional computer programs have ditched single precision and avoided the lousy algorithms. It's best not to mention Microsoft Excel at this point. That doesn't mean that you can't trip up one of these programs with a tricky data set, but in general, accuracy is not a serious concern. Several years ago, Statistica had a very aggressive advertising campaign that cast aspersions on many of its competitors. I think that there was also some pushback from people like Leland Wilkinson that held the claims to be unfair and unsupported. I wonder if this person has some residual memory of this campaign. If "accuracy" means something other than numerical accuracy, then you need to define what this really means. I could make some guesses, but that's dangerous. I would discourage a comparison of two statistical packages based on accuracy. The criteria that make more sense are: (1) ease of use, and (2) availability of advanced procedures. I have no experience with Statistica, but I'd be surprised based on what I have read about the program, that it would be considered vastly superior in either category. -- Steve Simon, Standard disclaimer. Looking for new friends/connections on Facebook and LinkedIn www.facebook.com/pmean, www.linkedin.com/in/pmean or become a fan of my newsletter, The Monthly Mean http://www.facebook.com/group.php?gid=302778306676 ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |