In most of social sciences research, 95% confidence interval is taken. My question is: If the research model is complex then can someone take 90% CI instead of 95%?
The level of CI is set by the researcher. So I do not think it would depreciate the value of your findings (considering the complexity of your model). The important thing is that you can make clear your reasons for defining a 0.10 alpha and what the consequences of this decision for the interpretation of your results (let the reader decide if they are OK with it). In some areas (health, for instance), it would not be accepted an alpha under 0.01 (99%). Although in the social sciences, despite typically adopting 95% (0.05), it is possible to accept a 90% (0.10) as well.
If you need some help with the arguments, you my see Hair et al (2009) in which he states that "[e]stablishing the significance level (alpha) denotes the chance the researcher is willing to take of being wrong about whether the estimated coefficient is different from zero. A value typically used is .05. As the researcher desires a smaller chance of being wrong and sets the significance level smaller (e.g., .01 or .001), the statistical test becomes more demanding. Increasing the significance level to a higher value (e.g., .10) allows for a larger chance of being wrong, but also makes it easier to conclude that the coefficient is different from zero".
Hazelrigg (2009) also may be helpful for you when he says that "as when setting confidence intervals, there is nothing sacrosanct or magical about these numbers, either Z or alpha. They are entirely conventional choices, and one is free to select a different number. Typically one begins with a value of alpha that is personally acceptable and that will be acceptable to one's audience (where 'acceptability' is defined in terms of managing errors of decision [...])".
I hope it may help you out somehow. Good luck!
Hair Jr., J.F., Black, W.C., Babin, B.J. and Anderson, R.E. (2009). Multivariate Data Analysis. 7th Edition, Prentice Hall, Upper Saddle River, NJ.
Hazelrigg, L. (2009). Inference. In: M. Hardy and A. Bryman (eds.). The Handbook of Data Analysis. Sage, London, UK.
Depends on level of significance you need...
Or risk your ready to bear...
The level of CI is set by the researcher. So I do not think it would depreciate the value of your findings (considering the complexity of your model). The important thing is that you can make clear your reasons for defining a 0.10 alpha and what the consequences of this decision for the interpretation of your results (let the reader decide if they are OK with it). In some areas (health, for instance), it would not be accepted an alpha under 0.01 (99%). Although in the social sciences, despite typically adopting 95% (0.05), it is possible to accept a 90% (0.10) as well.
If you need some help with the arguments, you my see Hair et al (2009) in which he states that "[e]stablishing the significance level (alpha) denotes the chance the researcher is willing to take of being wrong about whether the estimated coefficient is different from zero. A value typically used is .05. As the researcher desires a smaller chance of being wrong and sets the significance level smaller (e.g., .01 or .001), the statistical test becomes more demanding. Increasing the significance level to a higher value (e.g., .10) allows for a larger chance of being wrong, but also makes it easier to conclude that the coefficient is different from zero".
Hazelrigg (2009) also may be helpful for you when he says that "as when setting confidence intervals, there is nothing sacrosanct or magical about these numbers, either Z or alpha. They are entirely conventional choices, and one is free to select a different number. Typically one begins with a value of alpha that is personally acceptable and that will be acceptable to one's audience (where 'acceptability' is defined in terms of managing errors of decision [...])".
I hope it may help you out somehow. Good luck!
Hair Jr., J.F., Black, W.C., Babin, B.J. and Anderson, R.E. (2009). Multivariate Data Analysis. 7th Edition, Prentice Hall, Upper Saddle River, NJ.
Hazelrigg, L. (2009). Inference. In: M. Hardy and A. Bryman (eds.). The Handbook of Data Analysis. Sage, London, UK.
A .10 significance level is usually used with a small sample where you have low power to detect an effect.
Thanks Leandro for providing the references. My confusions are pretty much cleared now.
I think the answers provided cover it well. I hope along with the advice pertaining to smaller samples and complex designs. I tend to think that .90 is OK when you are doing original research where there are not a lot of previous studies. How big is your sample? Are these survey data? Please tell me a bit more about the complexity of your design.
My model has double mediation with 3-way moderation. An experimental study was done. The sample size for the experiment which tested mediation was 76 while the 2nd experiment was done to test moderation, the sample size for this was 77. All hypotheses of the study are not tested before. Specifically the mediation hypotheses.
The researcher is free on this issue. Traditionally 95% confidence interval use is widespread, but in social sciences, 90% confidence interval can also be used, especially in small sample sizes. Obviously, for a used estimation method, the confidence interval will decrease as well as the level of confidence.
You will find the right answer reading the following link:
https://www.surveymonkey.com/mp/sample-size/
- Badges
- Science topic
- Similar topics
- Social Science