todo
Differences
This shows you the differences between two versions of the page.
Next revision | |||
— | todo [2019/07/23 14:24] – external edit 127.0.0.1 | ||
---|---|---|---|
Line 1: | Line 1: | ||
+ | ===== To-Do List / Planned Features ===== | ||
+ | |||
+ | Below is a to-do list and a description of planned features that I would like to implement in the metafor package at some point in the future. This list is dynamic, incomplete, somewhat disorganized, | ||
+ | |||
+ | ==== Psychometric Meta-Analysis ==== | ||
+ | |||
+ | At some point, I plan on adding aspects of what is commonly called the Hunter and Schmidt (H&S) method for meta-analysis. To find out more about this approach, you could take a look at the book: | ||
+ | |||
+ | [[http:// | ||
+ | |||
+ | While the principles underlying this method can be applied to various outcome measures used in meta-analyses, | ||
+ | |||
+ | There are various aspects to the H&S method that deviate a bit from the statistical/ | ||
+ | |||
+ | - Write a completely different set of functions that exactly replicate the H&S method (and the corresponding software). | ||
+ | - Make it easy for users to specify the necessary information to correct correlations individually for various artifacts and then feed those corrected correlations into the modeling functions already available. Some tweaks/ | ||
+ | |||
+ | Option 1 would take a lot more work and would require separately maintaining that part of the package together with the rest. The advantage of course would be that users who really want to carry out the H&S method "by the letter" | ||
+ | |||
+ | Option 2 is what I am currently aiming for, because it is easier to implement and fits into the existing package framework. For example, to correct a correlation coefficient for measurement error, the user would need to specify, besides the coefficient and the sample size, also the reliability estimate(s) (for one or both variables). The rest is just the correction for attenuation. This option does not preclude that, at some point I (or anybody else) would write code to implement a line-by-line reimplementation of the H&S method. But that is currently not at the top of my to-do list. | ||
+ | |||
+ | Note that there is now a discussion of the H&S method under the tips section (see: [[tips: | ||
+ | |||
+ | ==== Bayesian Meta-Analysis ==== | ||
+ | |||
+ | I get this question quite a bit and the answer is: Yes, I definitely would like to implement methods for fitting meta-analytic models using Bayesian methods. In fact, I had developed code for this a long time ago (at the time, interfacing with [[http:// | ||
+ | |||
+ | There are a couple issues: | ||
+ | |||
+ | - As a start, I will implement this for models that can be fitted with the '' | ||
+ | - What choice of priors should be available? For the model coefficients (e.g., $\mu$ in a random-effects model), most people seem to use normal distributions (usually vague ones, with large variances). What if somebody is analyzing proportions or correlations? | ||
+ | - Even with ' | ||
+ | - How about other models besides the ' | ||
+ | - And then there is the '' | ||
+ | |||
+ | And I probably forgot some important issues ... | ||
+ | |||
+ | ==== Extending rma.glmm() to Multilevel/ | ||
+ | |||
+ | At the moment, the '' | ||
+ | |||
+ | ==== Alternative Graphing Systems ==== | ||
+ | |||
+ | Right now, all of the figures and graphs that can be created with the package are drawn using 'base graphics' | ||
+ | |||
+ | Fun fact: The grammar of graphics, which '' | ||
+ | |||
+ | At any rate, it would be nice to provide the option to create all figures and graphs in metafor using such alternative systems, especially given some of the limitations of the current approach (for a nice discussion of the underlying issue, see the article " | ||
+ | |||
+ | ==== Selection Models ==== | ||
+ | |||
+ | I am considering adding at least some version of a selection model to the package. But then again, there are the [[http:// | ||
+ | |||
+ | ==== Diagnostic Meta-Analysis Models ==== | ||
+ | |||
+ | The synthesis of diagnostic studies is of course a major application of meta-analytic methodology. Measures such as sensitivity and specificity can of course be easily meta-analyzed with the metafor package, but for things like the hierarchical summary receiver operating characteristic (HSROC) model and the binomial-normal version of the bivariate model (the normal-normal bivariate model can be handled with '' | ||
+ | |||
+ | ==== Miscellaneous Features and Things To Do ==== | ||
+ | |||
+ | Lots of other/ | ||
+ | |||
+ | * The '' | ||
+ | |||
+ | * Make '' | ||
+ | |||
+ | * The '' | ||
+ | |||
+ | * < | ||
+ | |||
+ | * Compute SEs of variance components (and covariances/ | ||
+ | |||
+ | * Also, make '' | ||
+ | |||
+ | * < | ||
+ | |||
+ | * < | ||
+ | |||
+ | * < | ||
+ | |||
+ | * The '' | ||
+ | |||
+ | * < | ||
+ | |||
+ | * < | ||
+ | |||
+ | * < | ||
+ | |||
+ | * Maybe allow optimization in '' | ||
+ | |||
+ | * Consider making use of some functionality from the [[http:// | ||
+ | |||
+ | * Allow ' | ||
+ | |||
+ | * Make metafor compatible with the [[https:// | ||
+ | |||
+ | * Add more analysis examples to this website, especially for some network meta-analyses. | ||
+ | |||
+ | * Write more tests. Really. I mean it. Lots of them. | ||
+ | |||
+ | ==== Some Additional Thoughts/ | ||
+ | |||
+ | * When I get the chance, I would like to spend some time examining to what extent using other math libraries speeds things up, especially for '' | ||
+ | |||
+ | * Alternatively, | ||
+ | |||
+ | * Various things are currently computed for '' | ||
todo.txt · Last modified: 2022/08/30 07:01 by Wolfgang Viechtbauer