---
tags: dexdex
---

# pairwise comparison

https://en.wikipedia.org/wiki/Pairwise_comparison

Robert thinks this a very efficient way to rank/assess lots of information

How could it be
leveraged in collective intelligence processes

Discussion on Metamaps slack on Pairwise Comparison
https://metamaps.slack.com/archives/C5Y2FB1QU/p1508203566000142

**All Our Ideas** has pairwise comparison

https://www.allourideas.org/planyc_example?guides=true

**Codigital** has pairwise comparison

https://app.codigital.com/p/planet?returnurl=https%3A%2F%2Fwww.codigital.com%2Fdashboard%2F


## Implementing pairwise comparison in a text-only medium, like chat

The following (needs improvement!) could represent an interaction between an automated process (an "Operator") and a person, to collect pairwise comparison results.

Obviously there are different things according to which you could be voting/selecting. You need to have the exact context to know what picking one or the other means. This would be configurable text per process. In this case, it's the "which would you rather?" (which isn't very goood context... "rather" for what?)

Here are two statements: which would you rather? :P
1) idea one
2) idea 2

reply with "1" or "2" to select one

@robertbest What would a good format in JSON look like, for representing pairwise comparison results?

## Some thoughts on the value of pairwise comparison.
* It is a way to take the elements from a large map with many nodes, and iterate through them with a human providing feedback on each comparison, and this can be integrated into a practice because you can just sit down for a chunk of time each day and answer some questions (like an ongoing iterative survey, that in some cases never ends, and is never intended to end)
* It gives you something very specific to focus on, rather than looking over a whole map, and being overwhelmed by all the potential relations/association. It also gives you a very particulat question to focus on while making a comparison. This means you could iterate over a map with one question in mind and reveal one layer of relationships, and then start fresh again later and ask a new question about the same set of nodes. If the question is framed well, it will make snap judgements as a glance easy (I have seen some pairwise comparison tools take into account how long you sat contemplating each pairing to try and use that metadata as also saying something meaningful)
* Often pairwise comparison is used to "rank" the set of nodes being evaluated, though this doesn't have to be the case. Different matchmaking algorithms could be used depending on the goal of any particular comparison. Maybe you are trying to only find the top 5 ranked of a set of 100 options, so once a certain option falls too low in the ranks, it would no longer make sense to include it in any future comparisons.
* Pairwise comparison also makes a lot of sense in the context of drawing graphs, particularly onese where all the possible connections still need to be drawn/discovered/etc... It is a systematic way to iterate over all the possible combinations (This desciption points at another possible mathmaking alorithm, i.e one that is exhuastive)
* Some examples of using pairwise comparison:
* If each node is a task, you could ask a simple question like, which of these two tasks should be done first?
* The pairwise comparison then lets you hone in on what needs done first... So it could let you rapidly prioritize a set of things.