Principal Components Analysis of Pilot Correlation Matrix

We then performed a principal components analysis on this matrix, rotating the space to find the axes of greatest variability. The PCA produced four eigenvalues greater than 1.0, with a significant gap between the fourth and fifth eigenvalues. The first four components account for 28, 11, 9, and 6 percent of the variability, respectively for a cumulative total of 54%. (See table 2.) To determine the meaning of the dimensions of the PCA, we examine the loadings of the attitude probes on each of the first four eigenvectors. (See table 3.)

The first principal component seems to encode something like ``discomfort with automation'' The probes with the strongest positive loadings on this component are NotReduceWorkload, SetupAndManage, and ButtonPusher. Other probes with positive loadings are FlyingSkills, CompanyPressure, Surprises, AltitudeBust, and MisunderstoodModes. The probes with negative loadings are WorksGreat, KnowMode, Freetomanage, FewerErrors, AheadOfAirplane, ConsultAnnunciator, AdequateTraining, and HelpDoJob. That is, this component has positive loadings on every one of the probes that express discomfort with automation and negative loadings on every probe that expresses comfort with automation. By their responses the pilots array themselves from comfortable to uncomfortable with automation on this dimension.

We have interpreted the second principal component as encoding an idea that is widespread in the culture of pilots, but which has received little formal recognition. The probes with the strongest positive loadings on this component are Surprises, MisunderstoodModes, and AheadOfAirplane. The strongest negative loading is for the probe KnowMode. The probes representing feeling like a button pusher and feeling that automation helps one do the job both have positive loadings on this component. At first glance this seems inconsistent. How could misunderstanding modes and keeping ahead of the airplane be aspects of the same thing? In interviews with pilots and observations of them flying as part of another study, pilots sometimes referred to the automation as ``magic''. An interpretation in which automation is believed to work well, but the pilot does not know why fits the loadings on this component.

The probes with the strongest positive loadings on the third principal component are ConsultAnnunciator and KnowMode and ButtonPusher. It also has positive loadings on CompanyPressure and AdequateTraining. The strongest, and only substantial negative loadings are for the probes Surprises and MisunderstoodModes. There is a mix of positive and negative attitudes with positive loadings on attitudes that represent conformance to procedure and negative loadings on attitudes that reflect confusion or misunderstanding. We interpret this dimension as encoding something like ``by the book'' or ``following orders''.

We were unable to find a coherent interpretation for principal component 4.