t-statistic or p value

[liliy702]

Hello,

I'm writing my thesis that as I found every references of data results are provided t-statistic value and using (*) to present p value due to my professor want me to replace t-statistic to p value number. so I'm not sure is the t-statistic value are the best way to presents the significant in smartPLS or p value?

Thank you for your help!!

[jmbecker]

Both are ok. It is more a matter of taste. Some people prefer t-values as they deliver more information (e.g., they are independent of the choice of whether you use one-tail or two-tail test). Other prefer p-values as they are easier to interpret and readers can directly assess the results without further calculations. However, p-values are based on the decision which tails you use and thus require more calculation if one wants to replicate the research and use a different approach (e.g., one-tail test).

Given the bootstrap approach in PLS, current recommendations also often include reporting the bias-corrected percentile confidence intervals. Usually, the results do not defer much from the t-value/p-value approach, but in some rare situations or where the decision is very close (e.g., p-values close to 0.05) the result can be slightly different whether an effect is significant or not. In those situations, current simulation studies show that the confidence interval approach is more reliable.

[liliy702]

thank you for your help!

[mpenn]

Does anyone have citations for articles showing that bias corrected confidence intervals are more reliable?

Thank you for your help,
mp

[jmbecker]

See also viewtopic.php?f=12&t=16221&p=27420#p27420