Exploratory factor analysis and reliability analysis with missing data: A simple method for SPSS users

Abstract

Abstract Abstract Missing data is a frequent problem for researchers conducting exploratory factor analysis (EFA) or reliability analysis. The SPSS FACTOR procedure allows users to select listwise deletion, pairwise deletion or mean substitution as a method for dealing with missing data. The shortcomings of these methods are well-known. Graham (2009) argues that a much better way to deal with missing data in this context is to use a matrix of expectation maximization (EM) covariances (or correlations) as input for the analysis. SPSS users who have the Missing Values Analysis add-on module can obtain vectors of EM means and standard deviations plus EM correlation and covariance matrices via the MVA procedure. But unfortunately, MVA has no /MATRIX subcommand, and therefore cannot write the EM correlations directly to a matrix dataset of the type needed as input to the FACTOR and RELIABILITY procedures. We describe two macros that (in conjunction with an intervening MVA command) carry out the data management steps needed to create two matrix datasets, one containing EM correlations and the other EM covariances. Either of those matrix datasets can then be used as input to the FACTOR procedure, and the EM correlations can also be used as input to RELIABILITY. We provide an example that illustrates the use of the two macros to generate the matrix datasets and how to use those datasets as input to the FACTOR and RELIABILITY procedures. We hope that this simple method for handling missing data will prove useful to both students and researchers who are conducting EFA or reliability analysis.

Corporate social responsibility and brand passion among consumers: Theory and evidence

“…It is apparent from the table that the EFA revealed four factors solutions with eigenvalues greater than 1.00. All the communalities were greater than the threshold of 0.60 (Weaver & Maxwell, 2014) and the items were loaded onto its factor. However, one item of brand trust (BT2) was deleted from the analysis because of significant cross‐loading (Siddiqui, 2015).…”

“…However, one item of brand trust (BT2) was deleted from the analysis because of significant cross‐loading (Siddiqui, 2015). Moreover, Kaiser–Meyer–Olkin was greater than the threshold of 0.70 and Bartlett's test was significant at p < .001, suggesting a good fit (Siddiqui, 2015; Weaver & Maxwell, 2014).…”

Environmental Factors of Knowledge Management Model for Implementation and Adoptation in the Construction Industry

“…Unidimensinality of the model is achieved by deleting all factor loading < 0.6 and convergent validity of the model is achieved using Average Variance Extracted (AVE) for every construct to be ≥ 0.5 as recommended by different researchers (Teo, 2010;Weaver & Maxwell, 2014) shown in Table 1 below. Composite reliability (CR) and internal consistency of the model were assessed and achieved with CR ≥ 0.6 as suggested by (Zainudin, 2014).…”

Kayıp veri atama yöntemlerinin faktörleştirme teknikleri üzerindeki etkisi

“…Akbaş ve Tavşancıl (2015) farklı kayıp veri oranları ve farklı örneklem büyüklüğü koşulları altında beş atama yönteminin temel bileşenler üzerindeki etkisini incelemiş ve çoklu atama ve beklenti maksimizasyonu yöntemlerinin çoğunlukla iyi sonuçlar verdiğini belirtmişlerdir. Sürekli veri setlerine dayalı olarak Weaver ve Maxwell (2014), temel eksenler yöntemine dayalı açımlayıcı faktör analizlerinde kayıp veriler için silme ve ortalama atama yöntemleri üzerine SPSS uygulama örnekleri gösterirken Josse ve Husson (2012) kayıp verilere çoklu atama ve beklenti maksimizasyonu yöntemlerinin temel bileşenler analizi üzerindeki etkisini incelemişlerdir. Lorenzo-seva ve van Ginkel (2016), çok boyutlu yapılarda farklı algoritmalar kullanan çoklu atama yöntemlerinin ağırlıklandırılmamış en küçük kareler yöntemine dayalı açımlayıcı faktör analizi üzerindeki etkisini araştırmışlardır.…”