Online Article

The number of samples needed to perform Exploratory Factor Analysis (EFA)

31st July 2024

Related topic: Quantitative research

Author: Ady Hameme N. A.

Exploratory Factor Analysis (EFA) is a widely used statistical method in psychology, social sciences, and behavioural research for uncovering the underlying structure of large variable sets. Determining the appropriate sample size is crucial when conducting EFA. This article examines recent research findings on the number of samples needed for effective EFA.

Traditionally, EFA has been viewed as requiring large sample sizes, with a commonly cited minimum of 50 samples. However, this guideline is not absolute and can vary based on factors such as the number of factors, variables, and factor loading levels.

Recent studies have challenged the necessity of large sample sizes for EFA. Research has shown that under certain conditions, such as high factor loadings, few factors, and many variables, EFA can produce reliable results with sample sizes well below 50. This finding is particularly relevant in behavioural research, where large samples can be difficult to obtain.

The presence of missing data further complicates sample size determination for EFA. Studies have found that deletion methods are ineffective for handling missing data in small samples, introducing severe bias in factor loading estimation and factor number extraction. Instead, techniques like predictive mean matching and 2-stage estimation have proven more reliable for EFA with small samples and missing data.

The type of measurement scales used in a study can also influence the required sample size for EFA. Different scales can affect the stability and reliability of the factor structure. Research has emphasised the need to carefully consider measurement scales when determining sample size, especially in clinical settings where large samples may be challenging to gather.

These findings have practical implications for researchers conducting EFA. While traditional guidelines suggest a minimum sample size of 50, researchers should consider their data's specific conditions, such as factor loadings, number of factors, and variables, which may allow for reliable results with smaller samples. Additionally, appropriate handling of missing data is crucial, with methods like predictive mean matching preferred over simple deletion. Finally, the measurement scales used should be taken into account when determining sample size.

In summary, the number of samples needed for EFA is not fixed. While 50 samples have been traditionally recommended as a minimum, recent research shows that reliable results can be obtained with smaller samples under certain conditions. Researchers should carefully consider their data's characteristics, including factor loadings, number of factors and variables, and the presence of missing data, to ensure reliable and valid EFA results, even with smaller sample sizes.

To provide more specific guidance, sample sizes for EFA should be carefully considered. A sample of 25 is generally insufficient, though it may suffice in rare circumstances with strong factor loadings and few variables. A sample of 50 is minimally acceptable but often lacks robustness, particularly with complex factor structures. A sample size of 100 typically offers adequate statistical power and reliability for most EFA applications. Interestingly, sample sizes exceeding 250 often surpass necessary thresholds and may lead to overanalysis or the detection of trivial effects.

Therefore, researchers should aim for an optimal sample size that balances statistical rigour with practical constraints. It's worth noting that prioritising a higher number of indicators per factor often proves more beneficial than simply increasing sample size. This approach can enhance factor stability and interpretability, even with moderate sample sizes.

Cite this article: Ady Hameme, N. A. (2024, July 31). The number of samples needed to perform exploratory factor analysis (EFA). Retrieved <insert month> <insert date>, <insert year>, from https://www.myadvrc.com/publications/article-18

< previous next >