what should be the range of t-value in regression analysis? in regression analysis on the basis of t- value coefficient s' significant level is measured along with the p value. is there any range to accept the coefficient as significant ? at 5% level t -value should be more than 2 ? Suppose t-value is 7 or 9 or12 or15?
Darshini -
Well, it is good to have the estimate of a coefficient's standard error to be smaller than the estimated coefficient itself, but there are many other factors to consider, and also remember that standard errors (unlike sigma) actually get smaller with larger sample sizes. But at least they are more useful and practical to interpret than a p-value.
[A single given p-value in general is not very useful.
https://www.researchgate.net/publication/262971440_Practical_Interpretation_of_Hypothesis_Tests_-_letter_to_the_editor_-_TAS]
To consider overall performance of different models, you could look at estimated variances of the prediction errors, or better, conduct a cross-validation study.
Cheers - Jim
Article Practical Interpretation of Hypothesis Tests - letter to the...
James R Knaub sir thanks for your reply. u mean that with larger sample size we get lower p-value. one important thing that std error value should be less than coefficient value. thanks sir.
y = a + .24b - 5.4c here coefficien is -5.4 what does indicate?
t valuue= 15.o +2.26 -5.90 here t value is -5.90 what does it mean ?
is there any roul that t value should be above 2(5%) to some value and coefficients should be less than 1 mean .69, .004 like wise except income value (coefficient).
I'm not clear on everything you are saying, but if you are looking for a rule, there really isn't one. It depends upon your goals, your particular application, sample size, effect size, and how your regressors might interact.
But even so, you can still say that the larger a t-statistic the better. If it were less than one, then you do not appear to have evidence (which may or may not be because of a too small sample) to say that your coefficient is anything but 'noise.' However, perhaps especially because of regressor interactions in multiple regression which make a measure on any individual regressor less meaningful, you cannot say that there is any particular cutoff point below which a t-statistic would indicate a useless regressor. - So, there isn't a rule.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis