Math of Jury Sizes

If we go by the assumption that a Jury is a representation of the public at large, then is 12 people statistically signficant?

When doing any scientific survey or poll, a sample of 12 people would be laughable. Particularly if the results are close to 50-50.

Is there any math done on this?

For example. We might go by some axioms such as, we should only convict someone if 2/3 of the public think they are guilty.

Then with that we can calculate the probability that a jury of 12 will wrongly convict or wrongly release a suspect.

For example if the proportional of the public that think the man is guilty is $$x$$. Then the probability that the jury will convict is if we convict if 2/3 of the Jury (8 or more men) say they are guilty is:

$$P(convict) = x^{12} + 12 (1-x)x^{11} + 66 (1-x)^2 x^{10} + 220 (1-x)^3 x^9 + 594 (1-x)^4 x^8$$

So for example if 3/4 of the public think the man is guilty he should be convicted. But the jury will convict 88% of the time and release on 12% of the time.

With the assumptions here is my question:

1. Assume public is effectively infinite
1. A man should be convicted if 2/3 of the public think he is guilty.
2. Given a number of jurors $$N$$ and what proportional of jurors $$f(N)$$ should we convict a man so that there is a 95% agreement with the public and the jury when 3/4 of the public think the man is guilty.
3. (Extra: Does this suggest an ideal jury size?)
• There is also the practical aspect in this case. Have you ever tried to instruct a group of, say, 1000 people to concentrate on one specific task? Especially if that task is "discuss amongst yourselves this difficult topic inside this closed room", that sounds like a recipe for chaos. Oct 17 '19 at 15:11
• @Arthur Yes, I've never like that aspect of Juries. Too much oportunity for bullying and such like. I think juries should not discuss the case with each other. But actually have a secret ballot. Unless this is to reflect wider society where people bully each other? Oct 17 '19 at 15:12
• Let's be careful to keep the discussion mathematical. This question is very prone to being closed for being opinion-driven.
– user694818
Oct 17 '19 at 15:16
• This is only tangentially related, but if we assume that jurors act strategically (and this is modeled in a very specific way), then it turns out that requiring unanimity can have some very counter-intuitive effects when it comes to the aggregation of the jurors' private information. This only gets worse with even larger juries. This is shown in Feddersen and Pesendorfer (1998). Oct 17 '19 at 15:26
• I suspect your second point should be something like "If the proportion of the public who thinks a person should be convicted is two-thirds or fewer then the probability the jury says guilty should be less than $5\%$", perhaps with some adjustment to the two-thirds or $5\%$ Oct 17 '19 at 17:16

Let's suppose you want people to be very unlikely to be found "guilty" when fewer than two-thirds of the public think they should be convicted, but very likely to be found "guilty" when three-quarters or more of the public think they should be convicted. This will lead to the need for large juries, for example

• if the proportion of the public who thinks an individual should be convicted is $$0.665$$ then the jury should say "guilty" with probability of $$5\%$$ or less
• if the proportion of the public who thinks an individual should be convicted is $$0.75$$ then the jury should say "guilty" with probability of $$95\%$$ or more
• the jury is a random sample of $$n$$ members of the public and at least $$g$$ of the jury have to vote "guilty" (without discussion) for "guilty" to be the collective verdict of the jury

then, using the binomial distribution, the following values of $$n$$ and $$g$$ will provide a satisfactory solution, and there will be more cases a jury sizes increase. The first case is optimal in the sense of a minimum jury size

  n   g
311 221
314 223
318 226
321 228
322 229
324 230
325 231
326 232
327 232
328 233
329 234
330 234
331 235
332 236
333 237
334 237
335 238
336 239
337 either 239 or 240
338 240
339 241
340 either 241 or 242


I suspect that this is not the size of jury that you were thinking about. If not, then you need to adjust the criteria in the first two bullet points