If you’ve used our coding form templates, you’ve likely noticed that several of the coding form pages are set up for adding rows or columns. In this post, we want to talk about when to create and use multiple columns on these pages. In particular, we want to demonstrate how MetaReviewer can facilitate coding a wide range of effect sizes from a limited set of columns. I also wanted an opportunity to share pictures of my pets.
Every column/row is a unique snowflake
We’ve set up the coding form templates to automatically have the option to add columns on the Sample Characteristics, Condition Characteristics, and Measures pages. We picked these pages since they describe the study factors that are most likely to come in multiples within a study (e.g., multiple outcome measures). These various factors will eventually be knitted together on your Effect Size page, allowing you to enter individual effect sizes for each unique sample/condition/measure combination. Importantly, the number of columns do not need to match across pages - you can reuse the same columns multiple times on the Effect Size page. For this reason, we recommend that you only make unique columns on these pages AND only make columns that you plan to use when identifying effect sizes.
Similarly, each row on your Effect Size page should contain information relating to a unique contrast in your data. These contrasts will be defined by the various inputs on the Effect Size page, including your responses on the Sample Characteristics, Condition Characteristics, and Measures pages.
Some examples (with pets!)
So how might this work in practice? As a simple example, let’s say I wanted to code a study of different training methods for my pets. In this situation, we have three samples: my pet dogs, my pet rats, and my pet turtles.
In this study, I wanted to test the effect of training them with food incentives (either meat or fruit) compared to simple praise. My goal was to train them to fetch and to sit.
From this design (and the ridiculous number of pets I have), I would code three sample columns,
three conditions columns,
and two measures columns.
This gives me a lot of different effect sizes I could code. For instance, let’s say I was interested in seeing how effective meat (compared to praise) was as a training tool for each group of pets on each outcome. This would give me six unique effect sizes, with the following combinations:
| Sample 1 | Sample 2 | Condition 1 | Condition 2 | Measure 1 | Measure 2 |
| Dogs | Dogs | Meat | Praise | Fetch | |
| Rats | Rats | Meat | Praise | Fetch | |
| Turtles | Turtles | Meat | Praise | Fetch | |
| Dogs | Dogs | Meat | Praise | Sit | |
| Rats | Rats | Meat | Praise | Sit | |
| Turtles | Turtles | Meat | Praise | Sit |
Each resulting row would give me an estimate of how much more effective meat was than praise for training my pets.
You do not have to fill out every column on the effect size page. In many cases, you will only have one outcome measure to record. That is alright – you can leave the second Measure column blank.
Let’s say that I was next interested in seeing which species of mammal was best at being trained to fetch. For this, I can identify a consistent condition and vary my sample inputs to create unique contrasts:
| Sample 1 | Sample 2 | Condition 1 | Condition 2 | Measure 1 | Measure 2 |
| Dogs | Rats | Meat | Fetch | ||
| Dogs | Rats | Fruit | Fetch | ||
| Dogs | Rats | Praise | Fetch |
This would result in three effect sizes, estimating how much faster (most likely) my dogs were at learning to fetch compared to my rats, with one estimate for each training method.
Finally, let's say I wanted to examine which treat got my turtles moving the fastest. I can set up a contrast that directly compares my two edible incentive conditions against each other.
| Sample 1 | Sample 2 | Condition 1 | Condition 2 | Measure 1 | Measure 2 |
| Turtles | Turtles | Meat | Fruit | Fetch | |
What if I don’t have effect sizes?
Not all syntheses will build toward coding effect sizes. In these instances, you may still want to code various combinations of samples, conditions, etc. We’re currently working on building out this functionality in the coding form templates for non-quantitative syntheses. The principles described above will still apply for those studies – stay tuned for those updates.
All the pets, all the contrasts
One of our goals with MetaReviewer was to reduce the amount of time you spend coding. A big part of that is only requiring you to identify study factors once and then allowing you to use them in multiple ways to identify effect sizes. In this example, I only needed to enter six columns of data before the effect size page. In this post alone, I’ve identified ten different effects sizes, each describing a unique pet-based phenomenon. There are many, many more unique effects I could define from these six columns (the turtle fetching questions alone would keep me going for a while). With MetaReviewer, you don’t have to spend time copying and pasting the same information about dogs, rats, or turtles across multiple rows of data. You simply need to identify the combinations you’d like, and MetaReviewer does the combining for you.
So, in summation, code as few columns as you have to; get as many pets as you can handle :)
