 All right, thanks everyone for coming out to my talk. So my name is Dan Noble. I'm a senior lecturer and future fellow at the Australian Astroniversity here in Canberra, Australia. So what I want to overview today very briefly and very quickly is the Orchard package that we developed a couple of years ago, and more specifically, I want to introduce you to version 2.0, which has a number of substantial improvements over 1.0. Now, given the time limitations, I'm not going to live code, but I provided you with a vignette of sorts to allow you to move through and understand Orchard's functionality. And I'll post the link to this vignette on the video. But basically, it'll tell you here how to install the Orchard package. It'll direct you to the GitHub page, which is hosting it because it's currently not on Cran. And it just sort of emphasizes the need to make sure you have an updated version of R because it'll be needed to play well with the YIN needs package for Orchard to work. OK, so what are Orchard plots? So Orchard plots are plots of meta regression models. They're force-like plots, to so to say. And they're specifically interested in predicting the meta and lift mean within levels of a particular moderator in a meta regression or multilevel meta regression model. So just to give you an idea of what Orchard plots are, I'll introduce you to this limb data set, which comes with Orchard. And this is a study done by Nat Lim, Al Senior, and Shinichi Nakagawa, looking at the correlation between offspring number and size and mother size offspring trade-offs. So we can fit a meta regression model normally using the metaphor package. And it's rma.mv function. Orchard requires you fit meta regression models with metaphor objects. It only takes metaphor objects, so you'll need to use metaphor. But that's OK because metaphor is an incredibly versatile, flexible, and very powerful meta analysis package. So if you fit that model and we're interested in trying to predict the meta and lift mean across this different file, you can generate an Orchard plot quite easily by basically feeding in the model itself, along with the specification of the mod argument, which tells Orchard plot which moderator you want to plot. You feed it the data and a group. The group is needed to specify the sample size for the groups within the brackets here. So this could be study often. And that can differentiate between k, which is the total number of effects. So from that, there are a whole bunch of different options here. You can transform it if it's correlation coefficient, or ZR to correlation coefficient, which is what's being done here. You've also got colorblind friendly plotting options, because we have so many levels I've suppressed here. And what you get then is a nice Orchard plot, which depicts the meta-analytic mean, these dots here. They're 95% confidence interval and the 95% prediction interval along with the raw data in each one of these levels of file up. These data are weighted by their sample size in this particular instance, but it could be precision. It just depends on what you're interested in plotting. Now, one of the nice features is the ability to subset. So you'll notice here, there is a lot of data, a lot of levels where there's very little data, where we can get around that, because the EMEANS package allows us to make predictions at certain levels. And so what I want to show you here is how you can subset the data to certain levels, and also how you can use another function called mode results to capture the model output in a table, which can be useful for publications. So another way to specify this is using the mode results. And here, mode results takes the exact same arguments that Orchard plot does on its own. But here, for this particular situation, it will capture the model results in a table. Again, we'll feed in the file information, but this time we have a new argument called at. And this allows us to make predictions at particular levels of moderators in the model. In this case, we want Cordata, Anthropona, and Mollusca, because those are the levels where we have most of the data. And not only then do we get a limb MR results as a table, we can feed that into Orchard plots to generate the Orchard plot itself, where we have the menalic means and confidence and prediction intervals for each one of the levels where we have sufficient amount of data. Of course, you can then use this mod table, which is held in the limb MR results object, to produce a nice table of those mean estimates and confidence intervals. Now, often when we have meta-aggression models we're interested in plotting continuous variables. This is less common in meta-analysis, but still reasonably common. If you think about requiring, or looking at publication biases, like time lag bias when you're interested in year. And we can easily capture that using the bubble plot function. Bubble plots then allow you to, they take the, again, the same arguments, but they allow you to plot a continuous moderator, in this case year, and you can have an interaction with other different categorical levels to understand how those moderators vary within those categories differently. So this is just an example of an orchard plot. And again, we've effectively just feed it in the same sort of arguments that orchard plot takes. Now orchard's great because when we rely on the EM needs package, it means that instead of a single moderator, multi-level meta-aggression model we can take multi-moderator progression levels and we can basically marginalize means across different levels of other moderators. So for example, if we turn to a different dataset, this is a fish dataset from Odead Al, which is published in 2019. They fit much more complicated meta-aggression models where they've got different trait types with varying degree Celsius differences between treatments and different experiments design. So here we can just fit all those moderators in this meta-aggression here. And you can see it can get quite complicated. This is the output from that model. But what we can do here is because we have those at and by arguments built into orchard plot, we can basically set the predictions up so that we predict say at five, 10 and 15 degree Celsius differences between the treatments. And we can then marginalize over all the other elements or moderators that are in that model, okay? And so this is basically an example of the kind of orchard plot you'd get where you're predicting the degree differences. So if you had a five or 15 degree Celsius difference between your two treatments, what would be the anticipated change in effect size magnitude in each of the different trait categories? Now, one thing that you'll notice often in meta-analysis, particularly with subgroups, is that the variability in those subgroups can vary quite a bit. And this violates the homogeneity variance assumption. We can actually relax that in metaphor models. Metaphor has a great option here by including a category or moderator within the random effects, which allows us to estimate depending on what level. In this case, it's a different residual variance across the different trait levels. And again, we can feed that model in. And here it will then recognize that this is a heterogeneous variance model in a plot, variance is differently across the groups, depending on what the best estimate of variability is in those groups itself. Now that's just a brief introduction to the package. Obviously I don't have a lot of time in 10 minutes, but there are a number of other features like I squared and R squared functions that allow you to calculate both point estimates and bootstrapped estimates of I squared and R squared, as well as submerged function, which allows us to take different model results, objects and merge them together to create a single orchard plot as well. If you wanna find out more, check out the vignette, which goes through quite a lot of detail. And I'm happy to take any kind of questions you might have about orchard 2.0. On that note, thanks for listening and hope you find it useful.