The development of a confidence interval-based importance–performance analysis by considering variability in analyzing service quality Hsin-Hung Wu a, * , Jiunn-I Shieh b a Department of Business Administration, National Changhua University of Education, No. 2 Shida Road, Changhua City, Changhua 500, Taiwan b Department of Information Science and Applications, Asia University, Taiwan article info Keywords: Importance–performance analysis Confidence interval Sampling error Variability Point estimate abstract The traditional importance–performance analysis (IPA) uses the mean ratings of importance and perfor- mance to construct a two-dimensional grid by identifying improvement opportunities and guiding stra- tegic planning efforts. The point estimates of importance and performance vary from sample to sample such that the numerical analyses are different based upon different samples. Thus, using point estimates for items might lead the management to make false decisions. This study integrates confidence intervals and IPA to reduce the variability which enables the decision maker much easier to identify the strengths and weaknesses based upon the sample of size used. Moreover, the assumptions of equal and unequal population variances for constructing confidence intervals are discussed. Ó 2008 Elsevier Ltd. All rights reserved. 1. Introduction Importance–performance analysis (IPA), a simple but effective technique, has been widely applied to study customer satisfaction expressed as a function of both expectations related to importance and performance (Eskildsen & Kristensen, 2006; Magal & Leven- burg, 2005; Martilla & James, 1977). A commonly seen IPA is a two-dimensional grid, depicted in Fig. 1, constructed by plotting mean ratings of performance and importance. Importance is la- beled as the x-axis, whereas performance is labeled as the y-axis. This four-quadrant matrix can be used to identify improvement opportunities as well as to guide strategic planning efforts (Daniels & Marion, 2006; Deng, Kuo, & Chen, 2008; Graf, Hemmasi, & Niel- sen, 1992; Hollenhorst & Gardner, 1994; Skok, Kophamel, & Rich- ardson, 2001). The meanings of these four quadrants in IPA are as follows: (Daniels & Marion, 2006; Shieh & Wu, 2007). Quadrant I has the characteristics of both high performance and importance, which indicates that the firm has been performing well to gain competi- tive advantage. Quadrant II has high performance but low priority. That is, the firm has overemphasized (possible overskill) the items located in this quadrant. The items falling in Quadrant III has the characteristics of both low performance and importance, which can be considered as the minor weakness. Finally, Quadrant IV has low performance but high importance. The area of ‘‘concen- trate here” suggests that any item falls in this quadrant requires immediate attention for improvement and is the major weakness. The inability to identify these strengths and weaknesses can threa- ten a firm’s place in the market and typically results in low cus- tomer satisfaction. To construct this four-quadrant matrix along with identifying the strengths and weaknesses of a firm, a survey must be con- ducted. Assume the effective sample of size for each item is n, and the mathematical expressions of typical survey results are de- picted in Table 1 where there are k items in the questionnaire, mean and STD are the sample average and sample standard devia- tion for each item, respectively. The y-axis line (performance) can be drawn as follows: first, the performance value for each item is computed by the average value from all respondents from the sur- vey. Thus, the performance value for item i is P i ¼ P n j¼1 y ij =n, where y ij is the value from 1 to 5 by Likert-type scale, where 1 and 5 rep- resent the very dissatisfactory and very satisfactory, respectively, for jth respondent in ith item for i = 1, 2, 3, ..., k and j = 1, 2, 3, ..., n. Finally, the y-axis line is determined by P ¼ P k i¼1 P i =k. For the x-axis line (importance), the discussions are as follows: the importance value for each item is calculated by the average va- lue from all respondent from the survey. Thus, the importance va- lue for item i is I i ¼ P n j¼1 x ij =n, where x ij is the value from 1 to 5 by Likert-type scale, where 1 and 5 represent the lowest importance and highest importance, respectively, for jth respondent in ith item for i = 1, 2, 3, ..., k and j = 1, 2, 3, ..., n. Finally, the x-axis line is built by I ¼ P k i¼1 I i =k. Each item can be plotted by a pair of values, i.e., means of importance and performance. For instance, for item i, the pair of values is (I i ,P i ). After plotting all items, a decision maker is easily to identify the strengths and weaknesses of a company. The decision to judge where an item is located is dependent only upon the means of importance and performance for a particular 0957-4174/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2008.08.055 * Corresponding author. E-mail addresses: hhwu@cc.ncue.edu.tw, drhhwu@yahoo.com.tw (H.-H. Wu). Expert Systems with Applications 36 (2009) 7040–7044 Contents lists available at ScienceDirect Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa