Reply With Quote
Check the standard error too. It's the second part that determines significance!
I'm carrying out an n-way Anova in SPSS. The dependent variable is salary and one of the independent variables is level of English. The Anova is telling me the sig. level is 0.94, i.e. not significant, yet the means are as follows -
None - 36.30
Basic - 46.67
Intermediate - 33.30
Advanced - 83.00
How can the differences possibly not be significant when the Advanced mean is double that of other groups? Sorry if I'm being thick, and thank you for any help!
Check the standard error too. It's the second part that determines significance!
[FONT=Fixedsys]
Visit : [/FONT][URL="http://www.statsravingmad.com/"]Site[/URL]|[URL="http://www.statsravingmad.com/blog"]Stats Blog[/URL]|[URL="http://parzakonis.info/"]Personal mini-site[/URL]
[FONT=Fixedsys]Contact : [/FONT][URL="http://www.statsravingmad.com/contact/"]Contact[/URL]
I'm sorry, I don't understand. Could you explain what you mean in really simple terms (i.e. for someone who isn't that bright!)? I can't see anything in my output that says standard error. I'm using SPSS by the way - did I need to select a post hoc test or something else before running the Anova? Also, how would I present the standard error statistic in my results? Thank you!
Ok.
q1. What is the Levene Stat. Is it <0.05?
q2. Maybe, it's the case that the "Robust Tests of Equality of Means" are more appropriate.
[FONT=Fixedsys]
Visit : [/FONT][URL="http://www.statsravingmad.com/"]Site[/URL]|[URL="http://www.statsravingmad.com/blog"]Stats Blog[/URL]|[URL="http://parzakonis.info/"]Personal mini-site[/URL]
[FONT=Fixedsys]Contact : [/FONT][URL="http://www.statsravingmad.com/contact/"]Contact[/URL]
Levene's test came out at greater than 0.05. Would this affect the significance level? I don't know what you mean by Robust Test of Equality of Means - do you mean the Brown-Forsythe/Welch tests? Sorry again!
Yes, that's what I mean. What is their significance?Originally Posted by minnie2
[FONT=Fixedsys]
Visit : [/FONT][URL="http://www.statsravingmad.com/"]Site[/URL]|[URL="http://www.statsravingmad.com/blog"]Stats Blog[/URL]|[URL="http://parzakonis.info/"]Personal mini-site[/URL]
[FONT=Fixedsys]Contact : [/FONT][URL="http://www.statsravingmad.com/contact/"]Contact[/URL]
I've tried a non-parametric test & Spearman's Rho for the same data and it's still coming out as not significant. Surely it can't be right, looking at the difference in means?
Oops sorry, crossed post. The significance is 2.58/2.57 respectively.
Do, you mean 0.258/0.257?
[FONT=Fixedsys]
Visit : [/FONT][URL="http://www.statsravingmad.com/"]Site[/URL]|[URL="http://www.statsravingmad.com/blog"]Stats Blog[/URL]|[URL="http://parzakonis.info/"]Personal mini-site[/URL]
[FONT=Fixedsys]Contact : [/FONT][URL="http://www.statsravingmad.com/contact/"]Contact[/URL]
Yes, apologies
What is the variance of each group ?
Hi, the variance between groups is as follows -
none - 2311.063
basic - 2635.264
intermediate - 917.379
advanced - 7294.500
So a considerable difference. Would the high degree of variance in the 'advanced' group be the cause of the fact the differences in means are not significant? Sorry if that doesn't make much sense!
Sorry, I mean the variance *within* each group, not between.
That's part the problem. You can see graphically in the attachment.
You said the dependent variable is salary. It cannot take negative values. Note that a non trivial portion of all those normal curves falls below zero. Your data are not normally distributed. You need to think again about your model. Start will plots of the data themselves and see what they look like. Perhaps take logs and look again.
Posting Permissions
|
|