I am doing time series analysis with 32 sample size.
But in doing estimations my results are such that I found perfect my variables sign at the cost of insignificance. If I correct my significance by change lags some of my variables appear with wrong sign.
So I come to know that we can reject ho of insignificance even we have p=0.20 I need some opinion on it. I appreciate if anyone provide some sort of reference.
I only know about fisher any neyman. I want atleast 2 more references With supportive explanation for my defense in thesis
Kindly guide. Thank in advance
Hello T.S.,
When people flip a coin to make a decision, they are implicitly accepting an error rate of 50% in the decision-making process. Such an error rate would not be very convincing for claiming support for a given conclusion in the scientific community, however (at least not for a single "trial" or "study"). So, what is an acceptable level of risk in decision-making? That depends on: (a) the consequences of being wrong; and (b) the audience (and their threshold for "chance" vs. "non-chance" outcomes). Remember as well, that your question is only addressing one way in which decision-making can go wrong, that of the Type 1 error (claiming a difference/relationship when none exists).
In behavioral science studies, most folks default to Type 1 risk levels of .05 (5%) or .01 (1%). However, there might be instances (so-called "compromise" scenarios) in which one accepts a higher Type 1 risk in order to lower Type 2 risk, when no changes can be made to sample size or target effect size. Exploratory or pilot studies might be instances in which such trade-offs are made.
Also, while a single hypothesis test might have what seems a high risk level (e.g., the coin flip), if a set of 10 such studies all show the same outcome, the collective risk associated with the claim that all 10 could be wrong becomes vanishingly small.
Good luck with your work!
Hello T.S.,
When people flip a coin to make a decision, they are implicitly accepting an error rate of 50% in the decision-making process. Such an error rate would not be very convincing for claiming support for a given conclusion in the scientific community, however (at least not for a single "trial" or "study"). So, what is an acceptable level of risk in decision-making? That depends on: (a) the consequences of being wrong; and (b) the audience (and their threshold for "chance" vs. "non-chance" outcomes). Remember as well, that your question is only addressing one way in which decision-making can go wrong, that of the Type 1 error (claiming a difference/relationship when none exists).
In behavioral science studies, most folks default to Type 1 risk levels of .05 (5%) or .01 (1%). However, there might be instances (so-called "compromise" scenarios) in which one accepts a higher Type 1 risk in order to lower Type 2 risk, when no changes can be made to sample size or target effect size. Exploratory or pilot studies might be instances in which such trade-offs are made.
Also, while a single hypothesis test might have what seems a high risk level (e.g., the coin flip), if a set of 10 such studies all show the same outcome, the collective risk associated with the claim that all 10 could be wrong becomes vanishingly small.
Good luck with your work!
Given your small sample size (btw how many degrees of freedom you have?) You could test the predictive ability of the model (using AIC for example) instead of testing coefficient being different from zero.
I am not aware of any study that permits significance level at 20%. The conventional and standard levels of significance are 1%, 5%, 10%. I must admit I may not have read wide enough. But I certainly know that will not be acceptable in the natural sciences.
- Badges
- Science topic
- Similar topics
- Correction