Hi, statistics question, confidence intervals and significance levels ?

[Gamewizard]

How do you choose a confidence interval and set the significance level ? for example CI= 95% and a=0.05 . I have been doing questions on t-statistics and ANOVA and i cannot do this part of the question, so i dont get correct results then.

Please help

[DeltaScience]

it depends on what you need, normally this 2 parameters determine how reliably is your study, this is because the confidence interval will determine how much of your population will be represented in your model, since you have an unknown parameter and only a sample from the whole population, you need to guarantee that your unknown parameter will be inside this sample, thats why you need to set an confidence interval level to see how much are you taking, its called An interval estimate for a population parameter.

 

Also the significance level will tell how big is your test, it represents a type of error, also called error type I, in wich it determines the posibilities for rejecting the null hypothesis when this one is true, so the smaller this error error the best, 0,05 is the standard, but ou can use 0,01, 0,5, it depends of your data.

 

if your data is not too accurate or you see there's much error then you need to adjust this parameters in order to generate a valid model, even though the realiability of your model will be lower.

 

I hope this is usefull. for further information you can check for this book, i think you can find it online: Applied Statistics and Probability for Engineers, the author is: Douglas C. Montgomery.

 

 

Greets

[Blahah]

It is usually standard, at least in the biosciences, to use a CI of 95% (a= 0.05). However, you should consider the test you are using.

 

I will give an example with ANOVA. Does your data violate any of the assumptions of the test? ANOVA assumes homogeneity of variance and normally distributed data. Let's say your data is not normally distributed (but only a bit off). You can make your ANOVA more reliable choosing a CI of 99% (a = 0.01) to reduce the possibility of an error due to violation of the test assumptions.

 

So, usually you would just go for 95% CI, but if you have a reason to think the test might not be ideal (assumptions slightly violated for example) then you could use a higher CI to make your test more trustworthy.