Effect size of ANOVA

#1
The ANOVA test I run shows significance as .045, can I use this p-value to reject the null hypothesis of there is no difference among 3 groups?
Can I just use this p-value as the effect size of this ANOVA?
I'm totally new in stats, I'm sorry if it's a silly question.
 

Karabiner

TS Contributor
#2
This is used to reject the null hypothesis, correct.

It is not an "effcts size". It is a statement about how probable the sample data (or more extrenne sample data) are,
if the null hypothesis is true.

Apart from that, it is of very limited use to try to calculate effect size measures in sample data, since effect sizes refer
to populations, not to samples.

With kind regards

Karabiner
 
#3
This is used to reject the null hypothesis, correct.

It is not an "effcts size". It is a statement about how probable the sample data (or more extrenne sample data) are,
if the null hypothesis is true.

Apart from that, it is of very limited use to try to calculate effect size measures in sample data, since effect sizes refer
to populations, not to samples.

With kind regards

Karabiner
Thank you so much for your reply Karabiner. So when I report the result of the ANOVA I run, I can just use the p-value (which is <.05 in my research) to state that it is significant that there is differences among the 3 groups that I tested? But I thought the point of testing sample is to predict the population, so if I don't address the effect size, how can I explain the meaning of the sample? sorry if it's a silly question again, I'm just really confused.
And I also attached the ANOVA result in where I tested the generation difference as the independent variable and non-Mandarin dialect proficiency as the dependent variable. Is it possible that you can kindly look it up? but it's also understandable if it's too much to ask.

Thank you loads!

Jade
 

Attachments

Karabiner

TS Contributor
#4
I am sorry, but with a frequentist approach, you cannot tell how large the effect is in the population.
You can only "predict" (or conclude) that the effect is not zero, i.e. that some means differ in the population.

Of course, if your sample sizes are very large, then the sampling error is small and the parameters which
you found in the sample are roughly representative of the population. But with small samples, there is
large sampling error. You could look at the 95% confidence intervals of your three group means to get
an idea of how (un-)reliable the estimates are.

From your results I would maybe doubt that the analysis makes very much sense. Do you want to make
inferential statements based on group sizes of just n=7 or even n=5?

With kind regards

Karabiner
 
#5
I am sorry, but with a frequentist approach, you cannot tell how large the effect is in the population.
You can only "predict" (or conclude) that the effect is not zero, i.e. that some means differ in the population.

Of course, if your sample sizes are very large, then the sampling error is small and the parameters which
you found in the sample are roughly representative of the population. But with small samples, there is
large sampling error. You could look at the 95% confidence intervals of your three group means to get
an idea of how (un-)reliable the estimates are.

From your results I would maybe doubt that the analysis makes very much sense. Do you want to make
inferential statements based on group sizes of just n=7 or even n=5?

With kind regards

Karabiner

Wow, thank you Karabiner, for helping me with such patience. Yea I know the sampling is really small, but the total participants are only 56 students, it's a very marginal group of people, so the number for each group is 5,7,44. Does this mean that I cannot run an ANOVA for it?
Also, I want to pay someone to help me and walk me through my quantitative analysis, will you possibly be interested in it? If you do, please kindly contact me via echoguizhou@gmail.com
I have the raw data and the potential inferential statements that I want to make, it's just the test running and reporting bit that I really have trouble to handle by myself.
 

Karabiner

TS Contributor
#6
Technically, you can run such an analysis. Personally, I would just not conclude much from results which are based on only half a dozen cases.

With very different sample sizes, it is important that the variances do not differ much between groups, because this could falsely increase or decrease the p-value. By default, you could perform the analysis of variance with a Brown-Forsythe or a Welch correction for unequal variances.

With regard to personal help, maybe you should do a little web search for statistical consultants which offer special prices for students?

With kind regards

Karabiner
 
#7
Technically, you can run such an analysis. Personally, I would just not conclude much from results which are based on only half a dozen cases.

With very different sample sizes, it is important that the variances do not differ much between groups, because this could falsely increase or decrease the p-value. By default, you could perform the analysis of variance with a Brown-Forsythe or a Welch correction for unequal variances.

With regard to personal help, maybe you should do a little web search for statistical consultants which offer special prices for students?

With kind regards

Karabiner

Thank you so much for your reply Karabiner. I may cancel this test as the sampling for each group is really small and not balanced.
Another question is (sorry for having so many questions)
I want to compare the number of languages the family speak and the number of language the child of the family speak, to kinda state that the multilingualism is declining within family.
Does this appear to you just a mean comparison? or I should do a T-test to state it?

Thank you a million.
 

Karabiner

TS Contributor
#8
Thank you so much for your reply Karabiner. I may cancel this test as the sampling for each group is really small and not balanced.
Maybe you should discuss this with your supervisor/instructor? I don't know what you need the analysis for.

I want to compare the number of languages the family speak and the number of language the child of the family speak, to kinda state that the multilingualism is declining within family.
If you have "family" as the unit of observation, and only 1 child from each family is considered, then you
could perform a correlational anaylsis, using the Peasron or the Spearman correlation coefficient. This will
tell you the degree of association between the 2 measurements.

With kind regards

Karabiner
 
#9
Maybe you should discuss this with your supervisor/instructor? I don't know what you need the analysis for.


If you have "family" as the unit of observation, and only 1 child from each family is considered, then you
could perform a correlational anaylsis, using the Peasron or the Spearman correlation coefficient. This will
tell you the degree of association between the 2 measurements.

With kind regards

Karabiner
I see, I will check it out as you said.
Thank you again for being so patient on helping me out.