Add vtype for RR/OR

[bwiernik]

I also documented the vtype options for RD, which were available but undocumented.

Changes:

1. Add vtype = "AV" for RR and OR measures following the existing structure for RD. The vtype = "AV" options will have very similar properties to the sample-size weighted estimator described by Bakbergenuly et al. (2019) and is much easier to implement in the existing code structure.

2. Add vtype = "UB" for RR and OR measures following the existing structure for RD by subtracting 1 from the total group sample sizes.

3. Document the vtype options for RR, OR, and RD. RD options were previously available, but undocumented.

Ref:
Bakbergenuly, I., Hoaglin, D. C., & Kulinskaya, E. (2019). Simulation study of estimating between-study variance and overall effect in meta-analysis of odds-ratios. arXiv. http://arxiv.org/abs/1902.07154

[bwiernik]

@wviechtb The failing checks don't seem to be related to these changes?

[wviechtb]

Yeah, Rcgmin and Rvmmin are archived, which is the problem. I currently have a large pending update of metafor sitting in my local repo that I need to finish up (which also touches escalc()), so to avoid any hassles with merging, I cannot pull this in right now. Will try to do so after the update. Not sure if you would have to fetch any updates coming from my side in the meantime. We will get this sorted out.

[wviechtb]

Okay, now had a chance to look at this. It's nice to add this, but UB for the log risk and odds ratio isn't quite right. For a single binomial proportion, one can easily derive the expected value of $\frac{p(1-p)}{n}$ which is $\left(\frac{n-1}{n}\right)\frac{\pi(1-\pi)}{n}$, which shows that $\frac{p(1-p)}{n-1}$ is unbiased (so for RD, we just do this for the first and second proportion). But this doesn't transfer to $\ln(p)$ or $\ln(p/(1-p))$. So I would suggest to remove UB for those measures.