Measurement Reliability in Psychology: Method Development, Infatuation with Cronbach’s Alpha Coefficient, and Recommendations for Appropriate Reliability Estimation
| Autor: | Josip Novak |
|---|---|
| Jazyk: | angličtina |
| Rok vydání: | 2020 |
| Předmět: |
Estimation
psihometrija pouzdanost mjerenja Cronbachov alfa koeficijent McDonaldov omega koeficijent teorija testova strukturalno modeliranje Social Psychology Psychometrics Measurement reliability Test theory Structural equation modeling BF1-990 Reliability engineering Clinical Psychology Cronbach's alpha Internal consistency psychometrics measurement reliability Cronbach’s alpha McDonald’s omega test theory structural equation modelling Psychology Applied Psychology Reliability (statistics) Mathematics |
| Zdroj: | Psihologijske teme Volume 29 Issue 2 Psychological Topics, Vol 29, Iss 2 (2020) |
| ISSN: | 1849-0395 1332-0742 |
| Popis: | Svrha je ovoga preglednog rada uputiti na bogatu povijest koncepta pouzdanosti mjerenja i strjelovit napredak u procjeni pouzdanosti koji se dogodio od 1940-ih, što je rezultiralo velikim brojem pristupa procjeni pouzdanosti. Na početku je rada naveden kratak povijesni pregled nastanka koncepta pouzdanosti i metoda procjene pouzdanosti. Potom su opisani koeficijenti unutarnje konzistencije prateći njihov kronološki razvoj. Kritizirana je primjena Cronbachova α koeficijenta bez provjere potrebnih pretpostavki te su obrazložene česta pogrešna shvaćanja vezana uz Cronbachov α koeficijent. Navedene su pretpostavke potrebne za opravdanu primjenu α koeficijenta, pristranosti procjene u slučaju nezadovoljenosti potrebnih pretpostavki, ispravan način izvještavanja o pouzdanosti, kao i moguće primjene koeficijenta u svrhe različite od procjene pouzdanosti. Međutim, u praksi su pretpostavke za opravdanu primjenu α koeficijenta vrlo rijetko zadovoljene i često je opravdana primjena drugih koeficijenata, poput λ2, λ4, ρNDG ili ω, za kojeg se pokazalo da je u većini uvjeta najmanje pristran. Ponuđene su smjernice za primjenu navedenih koeficijenata. Najnoviji je pristup procjeni pouzdanosti strukturalnim modeliranjem, koje omogućava jednostavnu provjeru navedenih pretpostavki i u kombinaciji s ω i ωH koeficijentima rezultira vjerodostojnom procjenom pouzdanosti, a opravdan je u slučaju kompleksnih mjernih modela. Naposljetku, istaknuti su konceptualni problemi vezani uz procjenu pouzdanosti te su predložene smjernice za daljnja istraživanja pouzdanosti mjerenja. The purpose of this review article is to indicate the abundant history of the concept of measurement reliability and steep improvement in reliability estimation since the 1940s, which produced a large number of approaches to its estimation. At the beginning of the article, a brief historical review of the emergence of the concept of measurement reliability together with methods of reliability estimation is given. Afterwards, different coefficients of internal consistency are described in chronological order. Application of Cronbach’s α coefficient without verifying necessary assumptions is criticized, and common misconceptions about α coefficient are elucidated. Necessary assumptions for use of Cronbach’s α coefficient, its biases in case of unfulfilled assumptions, best practice of reporting, and potential applications of the coefficient beside reliability estimation are outlined. However, necessary assumptions extremely rarely hold in practice and application of some other coefficients is often more appropriate, such as λ2, λ4, ρGLB or ω, the least biased one in most cases. Guidelines for use of mentioned coefficients are offered. The most recent approach to reliability estimation is by using structural equation modelling due to its ability to conveniently assess aforementioned assumptions and its synergy with ω and ωH coefficients, as well as its appropriateness in case of complex measurement models. Finally, conceptual issues on measurement reliability are discussed and guidelines for further research of measurement reliability are suggested. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|