Using Principal Component Analysis to Better Understand Behavioral Measures and their Effects
| Autor: | Jaime Arguello, Anita Crescenzi |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
Variables
Computer science media_common.quotation_subject Dimensionality reduction 05 social sciences Workload 02 engineering and technology Task (project management) Range (mathematics) 020204 information systems Perception Principal component analysis 0202 electrical engineering electronic engineering information engineering 0501 psychology and cognitive sciences 050107 human factors media_common Cognitive psychology |
| Zdroj: | ICTIR |
| Popis: | An important question in interactive IR research is: What do search behaviors tell us about specific task characteristics, post-task perceptions, and post-task outcomes such as knowledge gains? Much research has explored this question from different perspectives. A common approach is to consider a wide range of behavioral measures and examine their differences based on dependent variables of interest (e.g., post-task perceptions). In this paper, we use principal component analysis (PCA), a dimensionality reduction technique, to analyze behavioral measures captured during three previously published studies. Using PCA, we examine the underlying phenomena being captured by different behavioral measures, and we examine the influence of these phenomena on different outcomes related to participants' post-task perceptions (e.g., workload, difficulty, engagement, etc.). We argue (and show) that PCA can provide several benefits. First, it can help us understand the behavioral phenomena captured by different measures. Second, it can help us determine which measures are ambiguous or unambiguous with respect to the underlying phenomena being captured. Third, it can help us understand how behavioral phenomena (vs. individual measures) relate to searchers' perceptions of their search experience. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|