Require urgent help on ANOVA
Locked
 
|
|
|
Without a full understanding of your situation, my shoot-from-the-hip
reaction is:
You do not just want plain ANOVA. You should take a look at MANOVA DISCRIMINANT or flip the reasoning over and treat purchasing as the DV and try logistic and or Categorical regression. Art Kendall Social Research Consultants On 7/2/2010 2:55 PM, John Watson wrote: ===================== 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
Art Kendall
Social Research Consultants |
Re: Require urgent help on ANOVA
|
|
In reply to this post by John Watson-12
Is it perhaps as simple as the following?
Run a series of separate Anovas, one for each importance statement. (Analyse - General Linear Model - Univariate). Under Options, click Estimates of Effect Size. Partial eta-squared will be printed next to the significance level. When there is only one factor in a given analysis, partial eta-squared and eta-quared are the same thing: an estimate of the proportion of variance in the DV that is accounted for by that factor. You then rank the attributes in order of eta-squared. A couple of caveats:
Mike Griffiths Date: Fri, 2 Jul 2010 11:55:25 -0700 From: [hidden email] Subject: Require urgent help on ANOVA To: [hidden email]
Get a new e-mail account with Hotmail - Free. Sign-up now. |
|
|
In reply to this post by John Watson-12
John,
As another poster suggested, I think you should consider fitting a binary logistic regression model with "purchased product A" (yes/no) as the dependent variable and each of the attributes as an independent variable. This would tell you which of the attributes, if any, is significantly predicting the purchase of product A while taking into account the other attributes. You would also obtain the relative predictive strength of each attribute. Before fitting this type of model, you should confirm that you have met the assumptions. There's so much more that could be said on this topic, but without more information I will stop for now.
Ryan
On Fri, Jul 2, 2010 at 2:55 PM, John Watson <[hidden email]> wrote:
|
Re: Require urgent help on ANOVA
|
|
This certainly depends on what the questioner wants to do. Ryan's solution would be best if the questioner wants, as Ryan says, to look at each attribute taking into account the others. On the other hand, if John wants to consider each question independently, he would be better off with separate analyses.
A regression analysis (including logistic regression) will answer sophisticated questions, such as whether the answer to a given question adds more information on top of the answers to the other questions. For example, suppose that two of the questions were "How much do you like the appearance?" and "How much do you like the colour?". If those, and only those, questions are put into the analysis at the same time, and the "appearance" question is significant, it means that there are aspects of the appearance that add significantly to predicting whether the customer will buy the product, on top of that which is already predicted by knowing their opinion of the colour. (I am ignoring further complications, such as suppression.) Regression is also a good way of estimating the total predictability of all the questions, taken as a whole. And it would be possible to examine interactions, if one wanted to and if there were enough power. If the questioner just wants to take each question at face value, however, I remain of the opinion that he should do a separate analysis for each question and compare the effect sizes. He could do this with a series of logistic regressions, but surely a series of Anovas (details as in my earlier post) would be easier to interpret. I am happy to be challenged or enlarged on - that is what discussion groups are all about! Mike Griffiths Date: Sun, 4 Jul 2010 21:54:48 -0400 From: [hidden email] Subject: Re: Require urgent help on ANOVA To: [hidden email] John,
As another poster suggested, I think you should consider fitting a binary logistic regression model with "purchased product A" (yes/no) as the dependent variable and each of the attributes as an independent variable. This would tell you which of the attributes, if any, is significantly predicting the purchase of product A while taking into account the other attributes. You would also obtain the relative predictive strength of each attribute. Before fitting this type of model, you should confirm that you have met the assumptions. There's so much more that could be said on this topic, but without more information I will stop for now.
Ryan
On Fri, Jul 2, 2010 at 2:55 PM, John Watson <[hidden email]> wrote:
Get a new e-mail account with Hotmail - Free. Sign-up now. |
Re: Require urgent help on ANOVA
|
|
Hi,
Perhaps you can also consider C5, a data mining model in the family of decision tree. It gives you the output of the important attributes in ranking order, the prediction and also it gives you the profile of those purchasers and non-purchasers. It will state for example, a purchaser is someone who is 18 yrs old, earning 1200/mth and single as one of the rule. C5 can be found in PASW Modeler. I hope this helps. Regards Dorraj Oet Date: Mon, 5 Jul 2010 14:12:14 +0100 From: [hidden email] Subject: Re: Require urgent help on ANOVA To: [hidden email] This certainly depends on what the questioner wants to do. Ryan's solution would be best if the questioner wants, as Ryan says, to look at each attribute taking into account the others. On the other hand, if John wants to consider each question independently, he would be better off with separate analyses. A regression analysis (including logistic regression) will answer sophisticated questions, such as whether the answer to a given question adds more information on top of the answers to the other questions. For example, suppose that two of the questions were "How much do you like the appearance?" and "How much do you like the colour?". If those, and only those, questions are put into the analysis at the same time, and the "appearance" question is significant, it means that there are aspects of the appearance that add significantly to predicting whether the customer will buy the product, on top ! of that which is already predicted by knowing their opinion of the colour. (I am ignoring further complications, such as suppression.) Regression is also a good way of estimating the total predictability of all the questions, taken as a whole. And it would be possible to examine interactions, if one wanted to and if there were enough power. If the questioner just wants to take each question at face value, however, I remain of the opinion that he should do a separate analysis for each question and compare the effect sizes. He could do this with a series of logistic regressions, but surely a series of Anovas (details as in my earlier post) would be easier to interpret. I am happy to be challenged or enlarged on - that is what discussion groups are all about! Mike Griffiths Date: Sun, 4 Jul 2010 21:54:48 -0400 From: [hidden email] Subject: Re: Require urgent help on ANOVA To: [hidden email] John,
As another poster suggested, I think you should consider fitting a binary logistic regression model with "purchased product A" (yes/no) as the dependent variable and each of the attributes as an independent variable. This would tell you which of the attributes, if any, is significantly predicting the purchase of product A while taking into account the other attributes. You would also obtain the relative predictive strength of each attribute. Before fitting this type of model, you should confirm that you have met the assumptions. There's so much more that could be said on this topic, but without more information I will stop for now.
Ryan
On Fri, Jul 2, 2010 at 2:55 PM, John Watson <[hidden email]> wrote:
Get a new e-mail account with Hotmail - Free. Sign-up now. Hotmail: Powerful Free email with security by Microsoft. Get it now. |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |