Importance-performance analysis (IPA: Martilla & James, 1977) is based on a graphical representation of a set of items on a Cartesian diagram. Axes represent the importance (Y) and performance (X) levels of a set of aspects of a composite concept. The concept can be, for instance, the quality of a study programme and the aspects its possible components (i.e. teaching, organization, resource availability, etc.). In general, performance is measured with reference to satisfaction levels and importance either by surveying directly the concerned population or (indirectly) estimating the respondents' relevance of aspects. In particular, the aspects' importance can be estimated by means of (linear or logistic) regression of the set of aspects on the overall evaluation of the concept.
If the origin of the axes if shifted towards the middle of the scale, the resulting quadrants highlight the aspects to be improved and those to be ignored. Attention should be concentrated on the chart's quadrant where high importance cross low performance levels, so that efforts should be done to improve aspect performance until it joins at least the stated importance.
IPA is a simple and attractive technique. Its critical point is the choice of the axis origin, especially the Y-axis, because it determines whether aspects can be ignored or should be improved. The Y-axis origin may be the mean of the standardized regression coefficients that measure the relationships between the single aspects and the concept evaluation.
In our paper we prove that the mean of coefficients reduces as their variability increases. So, if we hypothesize a certain amount of unmeasured variability, we can define a “grey zone” between the coefficient mean and the “downed” mean within which coefficients have ignorable values (Boccuzzo et al., 2010).
We propose, too, an extension of IPA for the comparative analysis of subpopulations. The analysis is based on the contemporaneous representation of the aspects observed on different subpopulations over the same chart adopted for the whole-population analysis. This representation method practically adds a third dimension to the IPA graph. The contemporary view of the relative positions of the subpopulations may help in isolating disadvantaged subpopulations and guessing cause-and-effect relationships between subpopulation characteristics and aspects' position in the chart. Hence, we can guess proposals for orienting policy decisions and interventions.
An application to job quality analysis is introduced and commented.
Boccuzzo G., Fabbris L., Scarsi E. (2010) Non tutto l'oro luccica. Criticità dei lavori dei laureati identificate tramite Importance-Performance Analysis. In: FABBRIS L. (a cura di) Dal Bo' all'Agorà. Il capitale umano investito nel lavoro, Cleup, Padova: 95-138.
Martilla J., James J. (1977) Importance-Performance Analysis, Journal of Marketing, 41(1): 77-79
Keywords: Importance-performance analysis; Regression analysis; Stratification; Cause and effect analysis
Biography: Luigi Fabbris is full professor in Social Statistics at the Statistics Faculty, the University of Padua, Italy, where he lectures in Survey Methodology and Social Statistics.
He authored or co-authored more than 340 scientific publications, among which 54 volumes on: (i) data quality and sample survey design, (ii) multivariate analysis, (iii) computer-assisted interviewing; (iv) environmental, agricultural and epidemiological studies, (v) social uneasiness of family; (vi) women at work, (vii) job and labour market analysis, (viii) evaluation of social services, (ix) information systems, social indicators and statistical observatory design, (x) social integration of immigrants.