November 18, 201510 yr Quote We had a recent discussion on this topic. All forum members are invited to continue this discussion here. Vishwadeep Khatri Well, it all depends on how risky the error is. If Type II error is much more riskier of the two (Type I and type II) and you want to make sure sample size does not become too large, you can decide to go for higher alpha value (while keeping beta low). If it is easy to get large samples, you can choose to keep both errors low. Ashok Motwani One can use this example to explain when the risk can be taken by increasing alpha value. If you’re analyzing airplane engine failures, one must lower the probability of making a wrong decision and use a smaller value of alpha. On the other hand, if you're making paper airplanes, one might be willing to increase value of alpha and accept the higher risk of making the wrong decision. Comments are appreciated. Vishwadeep Khatri Well, it depends on the objective. An aircraft's supplier quality assurance team may go for a high alpha but low beta while doing inspections with limited samples for supplier approval. This is so because a bad part being accepted is riskier than a good part being rejected. Ashok Motwani Dear Vishwadeep, I completely agree with your response. My example was for comparison purpose..while using 2 different scenarios using airplane as the product. Comments are appreciated. Satish Karivedha Alpha, Beta which are also called as type I and Type II errors, and they are inversely proportional to each other. When you are looking high degree accuracy of Type 1 Error, you are giving good room for Type II error. Coming back to the question when to use very low value of alpha and high value alpha refers to your need of study. Low value of alpha means, you’re ok with high type II errors, high value of alpha means, your focus is more towards controlling type II errors. Practically both of them are errors, based your requirement study your trading off one of them. These errors can be controlled by increasing the sample size. Ashok Sharma Thank you sir... Chuck Gillis The alpha value can be set for any value, it all depends upon your risk tolerance. Sometimes we'll set it to 20% when looking for directional signals..other times we will set lower when it absolutely has to be spot on.. a minimal risk scenario. I always come back to the adage of how much risk are you willing to accept. Sushant Kaul Hi need one input regarding Minitab 16. I generally report few kpi's using run charts. Every month I have to manually update the data base to have desired result. Even though graphs & worksheet is saved under project format I can't just refresh & update data in columns. Please share your inputs.. Erik Laufer The way I typically approach alpha is as a proxy for the type of analysis that I'm performing. If I'm in an exploratory mode, my alpha will be set higher. Perhaps 0.10-0.20. This makes sense when processes are variable, first dabbling with a process, we haven't utilized SPC/Multi-Vari-COV analysis/Gage R&R-MSA; or employing screening-type DOEs. More factors make it through the sluice, but I tighten that down as I move into characterization and optimization activities, ratcheting down the alpha to less than 0.05. I use the quote, "you will miss 100% of the putts you leave short..." Why incorrectly eliminate a possible factor by an overzealous alpha that is not warranted early on? The other graphic that you can use are the scales of risk...high impact of a mistake translating to loss of life, and/or, millions of dollars? Drive Alpha low...low impact, I can live with a higher likelihood of a mistake...let Alpha drift higher... Regards, Erik Jason Bodnar, Ph.D. Don't forget about using simulation theory to determine if the desired alpha is actually achievable given your situation. Think of alpha as desired versus computed. Vishwadeep Khatri Hi Jason, great comment! I am starting another discussion string on this - How to determine if the desired risk is achievable. Kiran Varri In a NASA situation a 1% alpha n in case of a extremely low risk scenario, a 20% alpha...n in between is why the world goes with 5%... Jason Bodnar, Ph.D. Commonly in engineering settings, the type 1 and 2 risks are prescribed in industry standards such as ISTM and others. Outside of an industry standard, type 1 and 2 are typically chosen as a function of the risk associated with acting on the statistical analysis results. This is the hard part. What is the impact of the observed Type 1 and 2 errors on the business? How risky or conservative can you be? Sadly, there is no single path here. This is the fun in understanding the mechanics of the statistical theory of hypothesis testing.
Create an account or sign in to comment