Language: English Français

# Geeting the alpha and beta values

This question was posted the Coverage assessment forum area and has 4 replies.

### Mark Myatt

Consultant Epidemiologist

Frequent user

19 Nov 2014, 15:22

Is this real data or made-up data? You give inconsistent information since 67.5% (one "informed guess" at the prior mode) does not lie between 40% and 60% (other informed guesses). There is a conflict here that needs to be resolved through discussion and re-examination of the available evidence (and maybe collection of further evidence). A common "trick" is to take an average of a set of "informed guesses". You have "between 40[%] and 60[%]" and 67.5%. We might say that our best guess at the prior mode is: ``` (40 + 60 + 67.5) / 3 = 56% ``` You also to capture the variability in your belief. You could use something like: ``` 56 - 40 = 16% = +/- 16% ``` but that would make for a strong prior. If you are not sure of your prior (and it seems to me that you should be uncertain because you have a wide range of credible values ... 40% to 67.5%) then you can and should include this uncertainty in your definition of the prior. You might want to use (e.g.) +/- 25% and pick the alpha and beta from Table 4 on page 83 of the SQUEAC / SLEAC Technical Reference for a prior mode of 55% with +/- 25% uncertainty. This gives alpha = 19.1 and beta = 15.6. A key thing is that if you are uncertain then use a prior that reflects that uncertainty. You could use BayesSQUEAC or the formulae give on page 81 of the SQUEAC / SLEAC Technical Reference to find a suitable combination of alpha and beta. I find that Table 4 is usually good enough and much simpler and either use that to find alpha and beta or use it as starting point for exploration with BayesSQUEAC. If you need any more help with this then do not hesitate to post follow-up questions here. CMN offer training in SQUEAC. I hope this is of some help.

### Mark Myatt

Consultant Epidemiologist

Frequent user

20 Nov 2014, 09:23

Stanley, I though that the data may have been "made-up" because of the conflict between the prior mode from the barriers / boosters analysis and the range of values deemed credible by program staff and the assessment team. This is not to say that we will never see such conflicts. When we do they should provoke re-examination of the evidence. If conflicts cannot be resolved then the prior should reflect that. Remember that the prior reflects your best informed guess at what coverage will be and any uncertainty about that best informed guess. Sometimes we can have a lot of faith in our best informed guess. This will usually occur in very bad programs and very good programs. Most programs we see will be somewhere between those extremes and we will (e.g.) be less certain about our best informed guess. In other situations we may have a lot of faith in our best informed guess because we have coverage estimates from previous coverage assessments of from similar programs in the same state. Without divine knowledge we always work with informed guesses. We use the special term "estimate" to denote an informed guess. With the Bayesian approach used in SQUEAC stage 3 we build an informed guess and then try to improve it by adding more information (i.e. the likelihood survey data). If you need help that you can contact CMN who should be able to put you in touch with trained SQUEACers (or SQUEAC trainers) in your area. Also, don't forget that you can always ask for help on these forums.

### Jose Luis Alvarez Moran

Normal user

20 Nov 2014, 09:49

Dear Stanley, Following Mark´s comment, if you wish to be in touch with local teams doing SQUEACs in your area and look into local examples please do write to cmnproject@actionagainsthunger.org.uk and we will put you in contact. regards