mirror of
https://github.com/jackyzha0/quartz.git
synced 2025-12-24 05:14:06 -06:00
84 lines
2.5 KiB
Markdown
84 lines
2.5 KiB
Markdown
---
|
|
title: "assigning-participants"
|
|
aliases:
|
|
tags:
|
|
- info203
|
|
---
|
|
|
|
|
|
# Methods
|
|
## Between Subjects
|
|
The participants are split into equal groups, each which reviews one of the alternatives
|
|
## Within Subjects
|
|
The participants are split into two grops, one group review the first alternative first, the other review the seconds alternative first.
|
|
## Latin Square
|
|
|
|
|
|
order of each group is different
|
|
# Counterbalanced assignments
|
|
e.g., typing speed might affect interface usage
|
|
you can use a pretest to assign participants to that typing speed is roughly balanced
|
|
|
|
many techniques: key is to have equal chance of each participant in each group
|
|
|
|
## offline counterbalancing
|
|
pre-testing then forming matches pairs, which are split between groups]
|
|
|
|
## online counterbalancing
|
|
- of ther is no pre-test pick a threshold that is likely to be about the middle
|
|
- as they come in
|
|
|
|
- dont need to ensure even nubmer of high and low typers.
|
|
- do need to ensure the same number of high/low typers in a and b
|
|
## dangers
|
|
### regression:
|
|
- find heady coins
|
|
- first flip them all (pre-test)
|
|
- if they land heads more than half, call them heady
|
|
- "feed them a snack"
|
|
- does snacking increase the natural tendency of coins
|
|
|
|
|
|
|
|
both regress towards the mean
|
|
|
|
### how to avoid
|
|
if the pretest is used to counterbalance, and assignment is random, then the error goes away
|
|
|
|
|
|
# should every participant use every alternative
|
|
three major srategies
|
|
|
|
- **Within**
|
|
- everyone tries all options
|
|
- good when not fussed about learning/practise/exposure isssues
|
|
- **between participants**
|
|
- each person tries only only of the options
|
|
- requires more people, and more attention to fair assignment
|
|
- has the benefit that each participant is uncorrupted
|
|
- most common forthings like web studies
|
|
- **Counterbalancing**
|
|
-
|
|
|
|
# hawthorne effect
|
|
results are a result of the act of exmperimenting itself not as a resyult of the manupulations of the experiment
|
|
|
|
can be avoided with random assignment
|
|
|
|
# vaccum cleaner example
|
|
- manipulation
|
|
- vacuum type
|
|
- measures
|
|
- speed
|
|
- cleanliness
|
|
|
|
- between subjects design: assign half the participants to each type.
|
|
- worried about individual differences
|
|
- within subjects design: everyone uses both interfaces:
|
|
- worried about ordering effects
|
|
|
|
- half try one first, the other try the other first (counterbalancing)
|
|
- each of the tasks should be difference e.g., clean differnt buildings/rooms
|
|
|
|
individual differnces: go based of intuition of if it will make a difference.
|
|
Random assignment is importantg |