GZ: Quench data update
Since finishing the classifications for the GZ: Quench project, many of our volunteers have been analyzing that consensus data using the tools at tools.galaxyzoo.org. We made a few changes to the site earlier this week, and I’d like to describe them and talk about how it might affect your work on the project.
First, a quick reminder of how the data is presented. As most of you probably remember, the classification process on GZ: Quench (and all GZ projects since GZ2) is what we call a “decision tree”. We begin with a broad question on morphology (ie, “Is this galaxy smooth, or does it have features or a disk?”) for the volunteer to answer. We then ask more specific follow-up questions that depend on the previous answers. For example – if you said the galaxy doesn’t have any spiral arms, it doesn’t make sense for us to then ask you how many arms there are – it doesn’t apply to this galaxy! So, out of 11 potential questions covering galaxy morphology, a single classifier will only answer a subset (between 4 and 9) of them. Here’s a flowchart of the decision tree for GZ: Quench — it’s an interesting exercise to look at it and work out how many unique morphologies you could sort galaxies into by going through the tree.
So, why this discussion? When we added the data to the Tools website, we added a label in each category that gave the most common response to that question. For example, under “Arm tightness”, you could see that all galaxies were either “Tight”, “Medium”, or “Loose”. However, this is problematic when you’re trying to analyze data and compare different sets of galaxies. For smooth (or elliptical) galaxies, though, this arm classification is the result of very few votes (or even none) — they don’t represent the majority of classifications, and thus we really shouldn’t be including them when trying to compare what makes a medium-wound vs. a loosely-wound spiral.
The solution we’ve adopted has been to edit the data on Tools — questions whose answers don’t apply to the consensus morphology (eg, spiral arms in a smooth galaxy, or the roundness of a spiral) are now blank. This means that if you look at the average color or size of any of these morphology properties, you’re now truly comparing similar groups of objects (apples to apples). Including other galaxies in earlier samples likely introduced a significant amount of bias – the science team thinks that this will largely help to address that.
What does this mean for your analysis? Most of your old Dashboards and results should still work and remain valid results. For any work where you were analyzing morphological details (especially for spiral structure), though, we encourage you to revisit these and run them again on the new, filtered dataset. Please keep posting any questions you have on Talk, and we’ll answer them as soon as we can. Good luck!