Hi there,

I think is the other way around!!!! This is important!!! If your test statatistics are insignificant, they mean that they are not able to reject the null hypothesis.

Notice that if you cannot reject the null it does not mean that the null is true!!!

This is like a fugitive facing a trial. The innocence principle is the null hypothesis. The fugitive will be set free if there is no convincing evidence that he commited a crime. This situation may happen because the evidence against him is scarce, or because some lawyer are incompetent, or because the fugitive is innocent indeed!!! The problem is that you never know.

In statistical testing is the same: no rejection of the null does not mean the null is true...tough call here!!!

Hope it helps!

what is the p value?

I agree with Pablo up above because failure to reject the null implies that the null MAY BE true and that should be your conclusion

However, if your econometric result are insignificant you should point it in your research results/discussion. It does not mean that results must be significant at mostly 5% level of significance every time. There are many research done that showed insignificance but they are relevant because of that or because of some other reason. It depends on the research points. It depends on what you want to confirm, the significance or insignificance? In this case, if you are willing to proceed with your research,you should clearly point your results that are not significant even at the 10% level and to explain why it happens to be as such. If you would like to get a significant results and to support/confirm some hypothesis/theories than you should consider some other methodology approach, including data, variables etc. I hope it helps. Best

You can go ahead and explain your result but point it out in the discussion that it is not significant and provide justification for such finding - not being significant in your case.

All the best.

Dear Zegeye,

Provided that the model is correctly specified, yes the suggestion forwarded by Pablo is correct. However, how far are you confident the model is correctly specified.

That is why one need to conduct all the various diagnostic tests on the estimated model including normality, heteroskedasticity, serial correlation, functional form, and stability test using the plots of cumulative sum and cumulative sum of squares. If all these tests are ok, then you are cool to go.

Thank you

The insignificance could have several reasons:

1) wrong model

2) Insufficient specification: functional form, too many or too few explaining variables

3) statistical mistakes or breaks in the data or in factors that influence the relations (careful examination of the data and especially of outlayers necessary)

4) very different levels of the coefficients (e.g. if there are 3 independent exogenous variables fluctuating around a similar mean, but (according to theory) the coefficients of the first two are, say, more than ten times higher than that of the third, one could likely get an insignificant estimate for the third, often with a wrong sign.)

5) correlation between exogenous variables

By the way, I think, usual significance tests are rather weak. The “automatic” Null-hypothesis for an expected positive coefficient is against a negative value (Can one exclude with a high probability that the true coefficient is negative?) If the estimated coefficient is significant, then almost all econometricians take exactly that value (often with more than 5 digits) without thinking whether its level is realistic. And most of them do not care about the constant from which one can often get valuable information about the quality of the estimate. I think, if one specifies a model, one should also have some ideas which levels of the coefficients are reasonable and could formulate the Null-hypotheses accordingly.

I think it depends on what your hypothesis states. Significance level can only reflect whether the change in your dependent variable as a result of the change in the independent variable is significant I.e how significant is the impact of the dependent variable on the independent variable. There can be a significant change or an insignificant change. The signs of the coefficients are very important determinants of whether to accept or reject the hypothesis

@Toun, The impact is rather the other way round. That is, " how significant is the impact of the independent variable on the dependent variable".

Thank you.

Because a lone p-value is virtually meaningless, wouldn't you really want prediction intervals around the pi_i values? Variance estimates would be meaningful and more practical than dealing with p-values and thresholds. Sample size impacts both variances and p-values, but practical use and understanding of a threshold means doing a sensitivity analysis such as a type II error probability analysis, or considering effect size. Wouldn't looking at variances be a better approach? I am not an expert on logistic regression, and after looking at a couple of my books, and the internet just now, I did not see a good reference for prediction intervals here. Perhaps someone else can help with that. (Don't forget heteroscedasticity.)

Also, some kind of residual analysis (perhaps graphical residual analysis?) for fit would seem useful. Perhaps there is a cross-validation or other procedure to avoid overfitting.

H(a) = Logistic model explains data

H(0) = Logistics model does not explain data

The result shows that logistic model is not significant. it means that it does not explain the data. The null hypothesis (H(o)) cannot be rejected. Your alternative hypothesis (H(a)) is wrong. Your model is wrong. Try different model specification.

I have got very important ideas. I would like to appreciate all.

While Paul Louangrath above is generally correct, his work appears very confusing.

First, the Ho ought to be written first then followed by Ha. The Ho is what we believe throughout the study until its completion. Using a non-nested type does not communicate the material effectively.

How about the null hypothesis of no serial correlation of autoregressive order zero to 8 and the alternative hypothesis of serial correlation.