Industrial Management & Data Systems

Gain more insight from your PLS-SEM results: the importance-performance map analysis
Christian M. Ringle Marko Sarstedt
Article information:
To cite this document:
Christian M. Ringle Marko Sarstedt , (2016),"Gain more insight from your PLS-SEM results: the importance-performance
map analysis", Industrial Management & Data Systems, Vol. 116 Iss 9 pp. -
Permanent link to this document:
[Link]
Downloaded on: 22 September 2016, At: 02:41 (PT)
References: this document contains references to 0 other documents.
To copy this document: permissions@[Link]
The fulltext of this document has been downloaded 9 times since 2016*
Users who downloaded this article also downloaded:
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

(2016),"New developments in partial least squares path modeling", Industrial Management & Data Systems, Vol. 116
Iss 9 pp. -
(2016),"Mediation analysis in partial least squares path modeling: helping researchers discuss more sophisticated models",
Industrial Management & Data Systems, Vol. 116 Iss 9 pp. -
(2016),"A tutorial on the use of PLS path modeling in longitudinal studies", Industrial Management & Data Systems, Vol.
116 Iss 9 pp. -

Access to this document was granted through an Emerald subscription provided by emerald-srm:468524 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald for Authors service
information about how to choose which publication to write for and submission guidelines are available for all. Please
visit [Link]/authors for more information.
About Emerald [Link]
Emerald is a global publisher linking research and practice to the benefit of society. The company manages a portfolio of
more than 290 journals and over 2,350 books and book series volumes, as well as providing an extensive range of online
products and additional customer resources and services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committee on Publication
Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archive preservation.

*Related content and download information correct at time of download.

 Gain More Insight from Your PLS-SEM Results: The Importance-Performance Map Analysis

Introduction

Partial least squares structural equation modeling (PLS-SEM; Chin, 1998; Garson, 2014; Hair, Hult,

Ringle, & Sarstedt, 2016; Lohmöller, 1989; Rigdon, 2013; Tenenhaus, Esposito Vinzi, Chatelin, & Lauro,

2005; Wold, 1982) is a variance-based method to estimate path models with latent variables. The PLS-
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

SEM approach is particularly useful when the study focus is on the analysis of a certain target construct’s

key sources of explanation. For example, the technology acceptance model (TAM; Davis, 1989; Davis,

Bagozzi, & Warshaw, 1989) and its various extensions, such as the unified theory of acceptance and use

of technology (UTAUT; Venkatesh, Morris, Davis, & Davis, 2003), are popular models for PLS-SEM

applications in management information systems research. In the marketing field, the American Customer

Satisfaction Index (ACSI) model (Anderson & Fornell, 2000; Fornell, Johnson, Anderson, Cha, & Bryant,

1996) is another widespread PLS-SEM application. PLS-SEM enjoys rapidly increasing usage in various

business disciplines, such as accounting (Lee, Petter, Fayard, & Robinson, 2011), family business

(Sarstedt, Ringle, Smith, Reams, & Hair, 2014), international business (Richter, Sinkovics, Ringle, &

Schlägel, 2015), management information systems (Ringle, Sarstedt, & Straub, 2012), marketing (Hair,

Sarstedt, Ringle, & Mena, 2012), operations management (Peng & Lai, 2012), strategic management

(Hair, Sarstedt, Pieper, & Ringle, 2012), and tourism research (do Valle & Assaker, 2015).

The purpose of this article is to explain and illustrate the use of the importance-performance map

analysis (IPMA; also called importance-performance matrix, impact-performance map, or priority map

analysis), a useful analysis approach in PLS-SEM that extends the standard results reporting of path

coefficient estimates by adding a dimension that considers the average values of the latent variable scores.

More precisely, the IPMA contrasts the total effects, representing the predecessor constructs’ importance

in shaping a certain target construct, with their average latent variable scores indicating their performance

(Fornell et al., 1996; Martilla & James, 1977; Slack, 1994). The goal is to identify predecessors that have a

-1-
 relatively high importance for the target construct (i.e., those that have a strong total effect), but also have

a relatively low performance (i.e., low average latent variable scores).

Illustrative Example

To illustrate the concept of an IPMA, consider the PLS path model in Figure 1 with four constructs

Y1 to Y4. In this PLS path model, Y4 represents the final target variable, directly predicted by Y1, Y2, and Y3.
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

Furthermore, Y1 and Y2 have indirect effects on Y4 via Y3. Adding the predecessor constructs’ direct and

indirect effects yields their total effects on Y4, which represent the importance dimension in the IPMA. In

contrast, these constructs’ average latent variable scores represent their performance, in the sense that high

values indicate a greater performance.

Figure 1: IPMA Model

The IPMA combines these two aspects graphically by contrasting the (unstandardized) total effects

on the x-axis with the latent variable scores, rescaled on a range from 0 to 100, on the y-axis. The result is

a chart such as in Figure 2. For the results interpretation, we focus on constructs in the lower right area of

the importance-performance map. These constructs have a high importance for the target construct, but

show a low performance. Consequently, there is a particularly high potential to improve the performance
-2-
 of the constructs positioned in this area. Conversely, the performance improvements priority of constructs

with lower importance relative to the others, is less pronounced.

Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

Figure 2: Importance-performance Map of the Target Construct Y4

In Figure 2, Y1 is particularly important to explain the target construct Y4. More precisely, a one-unit

point increase in Y1’s performance increases the performance of Y4 by the value of Y1’s total effect on Y4,

which is 0.84 (ceteris paribus). Since the performance of Y1 is relatively low, there is substantial room for

improvement, making the aspect underlying this construct particularly relevant for managerial actions.

While this introductory example shows an IPMA on the construct level, the analysis can also be run on the

indicator level. In this case, individual data points in the importance-performance map are derived from

indicator mean values and their total effect on a particular target construct.

The IPMA Procedure

PLS-SEM studies that draw on IPMA results offer important insights into the role of antecedent

constructs and their relevance for managerial actions (e.g., Grønholdt, Martensen, Jørgensen, & Jensen,

-3-
 2015; Höck, Ringle, & Sarstedt, 2010; Kristensen, Martensen, & Grønholdt, 2000; Martensen, Dahlgaard,

Park‐Dahlgaard, & Grønholdt, 2007; Martensen & Grønholdt, 2010). The IPMA also becomes

particularly useful when contrasting PLS-SEM results from a multigroup analysis (Hair et al., 2016;

Sarstedt, Henseler, & Ringle, 2011), as several studies illustrate (Rigdon, Ringle, Sarstedt, & Gudergan,

2011; Schloderer, Sarstedt, & Ringle, 2014; Völckner, Sattler, Hennig-Thurau, & Ringle, 2010). However,

to date, no comprehensive tutorial highlights the requirements for using the method, or offers a step-by-
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

step introduction to its use. Against this background, this article presents a state-of-the-art review and

detailed explanation of the IPMA. A case study on the TAM illustrates the analysis, using the SmartPLS 3

(Ringle, Wende, & Becker, 2015) software. The article explains an IPMA by following the five-step

procedure as shown in Figure 3.

Figure 3: Steps of the Importance-Performance Map Analysis

When using an IPMA, the first step involves checking if the requirements for carrying out the

analysis have been fulfilled (Step 1). The analysis proceeds with the computation of the latent variables’

performance values (Step 2) and their importance values (Step 3). The importance-performance map

creation for a selected target construct is based on these results (Step 4). Finally, the IPMA can be

-4-
 extended on the indicator level to obtain more specific information on the most effective managerial

actions (Step 5). The following sections explain each step in greater detail.

Step 1: Requirements Check

IPMA applications have to meet three requirements. First, the rescaling of the latent variable scores

on a range from 0 to 100 requires all indicators in the PLS path model to use a metric or quasi-metric scale
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

(Sarstedt & Mooi, 2014). Second, all the indicator coding must have the same scale direction. The

minimum value of an indicator must represent the worst outcome and the maximum value must represent

the best outcome of an indicator. Otherwise, we cannot conclude that higher latent variable scores

represent better performance. If the indicator coding has a different direction compared to the other

indicators in the measurement model (i.e., a high value represents a negative outcome), we must re-scale

the indicator. In this case, the indicator coding needs to be changed by reversing the scale (e.g., on a 5-

point scale, 5 becomes 1 and 1 becomes 5, 4 becomes 2 and 2 becomes 4, and 3 remains unchanged).

Third, regardless of the measurement model being formatively or reflectively specified, the outer weights

estimates must be positive. If the outer weights are negative, the latent variable scores will not fall within

the 0 to 100 range, but would, for example, be between -5 and 95. Note that there are different reasons for

(unexpected) negative outer weights. If an outer weight is negative and significant, the researcher should

inspect the indicator and its scale. It may have another direction compared to the other indicators in the

measurement model, which requires reversing the scale. In case of nonsignificant outer weights (with

negative signs), the researcher may consider removing those indicators. Finally, negative outer weights

might be a result of high indicator collinearity. For example, variance inflation factor (VIF) values of 5

and higher indicate a potential collinearity problem (Hair et al., 2016). In this case, the researcher may

also consider removing indicators. However removing indicators from measurement models involves

some additional considerations as explained by Hair et al. (2016; see Chapter 5) in more detail.

While not being a formal requirement for running an IPMA, researchers should carefully consider

PLS path model set-ups that favor IPMA use on the indicator level. When a latent variable of high priority

-5-
 for a specific target construct is identified, it is particularly advantageous to further analyze this

predecessor construct’s measurement model on the indicator level. Such an assessment is particularly

useful when the measurement model is specified as formative—as is the case in our sample model in

Figure 1. In this case, the indicators describe aspects that shape the corresponding construct, while their

weights indicate each aspect’s importance in this respect. Therefore, aspects underlying indicators with

high weights should be given more attention to identify managerial actions aiming to improve the target
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

construct’s performance—see Höck et al. (2010) for an application. Note that the IPMA can be applied on

any kind of PLS path model, regardless of whether the researcher specifies latent variables’ measurement

models as formative or reflective. The IPMA builds on the outer weights—as explained in more detail in

the subsequent sections—and PLS-SEM always provides outer weights estimates, also when a

measurement model is specified as reflective. In this context, it is important to note that the distinction

between reflectively and formatively specified constructs refers to the ways how researchers develop

proxies for conceptual variables and the resulting measurement approaches. While PLS-SEM readily

processes reflectively and formatively specified constructs, it does so by linearly combining indicators to

form composite variables. These composite variables are treated as proxies of the concepts under

investigation (Rigdon, 2012, 2014) and serve as input for the IPMA in Step 2 (Figure 3). To represent

formative measurement models, PLS-SEM draws on composite indicatorsas opposed to causal

indicators, which fully form the latent variable without an error term on the construct level (Bollen &

Bauldry, 2011; Bollen & Diamantopoulos, in press). At the same time, PLS-SEM only approximates

measures in reflective measurement models that draw on a factor model logic. While the “bias” that PLS-

SEM produces when estimating common factor models is very smallprovided that measurement models

meet minimum recommended standards in terms of the number of indicators and indicator

loadingsrecent research has also introduced the consistent PLS (PLSc) approach that handles common

factor model-based measures without limitations (Dijkstra & Henseler, 2015). Acknowledging the proxy

character of the method, the following sections refer to the common denotation of reflective and formative

measurement models and their standard treatment in PLS-SEM analyses.

-6-
 Step 2: Computation of the Performance Values

The indicator data determines the latent variable scores and, thus, their performance. Similarly,

when conducting an IPMA on the indicator level, the mean value of an indicator represents its average

performance. When computing average values on the construct or indicator level, it is important to

remember that indicators may be measured on different scales. For examples, some indicators may use a
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

scale with values from 1 to 5, while others use a scale with values from 1 to 7, or from 1 to 9. To facilitate

the interpretation and comparison of performance levels, the IPMA rescales indicator scores on a range

between 0 and 100, with 0 representing the lowest and 100 representing the highest performance. Since

most researchers are familiar with interpreting percentage values, this kind of performance scale is easy to

understand. The rescaling of an observation j with respect to indicator i proceeds via

E[ xij ] − min[ xi ]
xijrescaled = ⋅100 , (1)
max[ xi ] − min[ xi ]

where xi is the i-th indicator in the PLS path model; E[.] represents indicator i’s actual score of

respondent j, min[.] and max[.] represent the indicator’s minimum and maximum value. It is important to

note that the minimum and maximum values refer to the potential values on a certain scale (e.g., 1 and 5

on a 1 to 5 scale) and not the minimum and maximum values of the actual responses (e.g., 2 and 4 on a 1

to 5 scale). Hence, if respondents use lowest actual response is 2 but the scale has a minimum value of 1,

it is mandatory to use the 1 as minimum value for rescaling. For example, according to this formula, a

value of 4 on a 1 to 5 scale becomes (4 – 1) / (5 – 1) · 100 = 75 while a 4 on a 1 to 7 scale becomes (4 – 1)

/ (7 – 1) · 100 = 50. All data points used for estimating the PLS path model are rescaled this way.

Table 1 shows an excerpt of the original indicator data (n = 300) used to estimate the sample

model from Figure 1. All indicators are measured on a scale from 1 to 5.

-7-
 Case x11 x12 x13 x14 x21 x22 x23 x24 x31 x32 x33 x34 x41 x42 x43 x44

1 5 2 1 5 3 3 4 5 5 2 4 3 1 2 4 3

2 4 5 2 3 5 4 5 3 4 1 1 2 3 5 4 3

3 4 1 3 4 5 3 4 5 5 3 5 3 3 5 3 1

4 1 3 2 1 3 2 2 4 1 4 1 3 5 1 4 4

5 3 2 1 3 5 1 2 3 5 2 2 3 1 3 4 5

… … … … … … … … … … … … … … … … …
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

299 2 4 3 4 4 3 3 4 4 2 2 1 2 5 4 3

300 2 4 3 1 5 2 3 5 1 1 4 3 4 2 1 5

Mean
4.2 4.1 3.4 2.3 3.6 4.5 3.4 3.1 3.4 4.7 4.4 4.6 3.4 4.2 4.5 4.2
Value

Table 1: Original Indicator Data

Table 2 shows the indicator data from Table 1, rescaled on a range from 0 to 100, which serve as

input for the computation of the rescaled latent variable scores. In addition, the mean values of the

rescaled indicators represent their performance values (e.g., 79 for indictor x11 and 77.5 for indicator x12),

which are later used for the IPMA on the indicator level.

-8-
 Case x11 x12 x13 x14 x21 x22 x23 x24 x31 x32 x33 x34 x41 x42 x43 x44

1 100 25 0 100 100 25 75 50 50 50 75 100 0 25 75 50

2 75 100 25 50 75 0 0 25 100 75 100 50 50 100 75 50

3 75 0 50 75 100 50 100 50 100 50 75 100 50 100 50 0

4 0 50 25 0 0 75 0 50 50 25 25 75 100 0 75 75

5 50 25 0 50 100 25 25 50 100 0 25 50 0 50 75 100

… … … … … … … … … … … … … … … … …
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

299 25 75 50 75 75 25 25 0 75 50 50 75 25 100 75 50

300 25 75 50 0 0 0 75 50 100 25 50 100 75 25 0 100

Mean
79.0 77.5 59.5 33.5 66.0 87.5 59.5 53.0 59.5 91.5 85.5 90.0 59.5 79.0 88.5 79.0
Value

Table 2: Rescaled Indicator Data

The rescaled latent variable scores are a linear combination of the rescaled indicator data and the

rescaled outer weights—whether the measurement model of a latent variable is reflective (i.e., Y4) or

formative (i.e., Y1, Y2, and Y3). To obtain the rescaled weights, we must first compute the unstandardized

weights by dividing the standardized weights by the standard deviation of its respective indicator. While

the standardized outer weights originate from the standard PLS path model estimation, the estimation of

each indicator’s standard deviation is based on the original indicator data. For example, if x11 has a

standardized weight of 0.2 and a standard deviation of 1.619, the resulting unstandardized weight is 0.124.

Table 3 shows the (standardized and unstandardized) indicator weights along with the indicators’ standard

deviations with regard to our sample model in Figure 1.

-9-
 Latent Standardized Standard Deviation Unstandardized Rescaled
Indicator
Variable Outer Weights of the Indicators Outer Weights Outer Weights

x11 0.2 1.619 0.124 0.139

Y1 x12 0.3 1.789 0.168 0.189

x13 0.4 2.099 0.191 0.215

x14 0.5 1.231 0.406 0.457

x21 0.1 2.099 0.048 0.114

Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

Y2 x22 0.1 1.101 0.091 0.218

x23 0.4 3.357 0.119 0.285

x24 0.4 2.504 0.160 0.383

x31 0.1 1.762 0.057 0.136

Y3 x32 0.1 1.744 0.057 0.137

x33 0.4 2.270 0.176 0.422

x34 0.4 2.653 0.151 0.361

x41 0.3 2.164 0.139 0.186

Y4 x42 0.3 1.874 0.160 0.215

x43 0.3 1.413 0.212 0.286

x44 0.3 1.291 0.232 0.313

Table 3: Computation of Unstandardized and Rescaled Outer Weights

Finally, we rescale the unstandardized outer weights so that their sum equals one per measurement

model. For this purpose, we need to divide each indicator’s unstandardized weight (e.g., 0.124 for

indicator x11) by the sum of the unstandardized weights of all the indicators that belong to the same

measurement model. For Y1, the sum of all the unstandardized indicator weights is 0.124 + 0.168 + 0.191

+ 0.406 = 0.889. Therefore, for indicator x11, we obtain the unstandardized and rescaled outer weight of

0.139 after dividing 0.124 by 0.889. The final column in Table 3 shows the results of the rescaled outer

weights.

- 10 -
 In the next step, the IPMA uses the rescaled indicator data (Table 2) and the rescaled outer weights

(Table 3) to compute the rescaled latent variable scores by means of simple linear combinations. For

example, the first data point in the vector of Y1’s scores is:

100 · 0.139 + 25 · 0.189 + 0 · 0.215 + 100 · 0.457 ≈ 64.3. (2)

Table 4 shows the resulting latent variable scores along with their mean values. In our example, Y1

has a mean value (i.e., performance) of 53.7, Y2 of 85.6, Y3 of 61.8, and Y4 of 78.1. These results serve as
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

input for the importance-performance map’s performance dimension.

Rescaled Latent Variable Scores

Y1 Y2 Y3 Y4

1 64.3 76.3 63.3 42.5

2 57.6 75.4 18.2 67.9

3 55.5 82.0 76.5 45.1

4 14.8 47.0 26.9 63.5

5 34.5 37.7 43.3 63.5

… … … … …

299 62.7 62.4 22.9 63.3

300 28.4 69.4 47.1 50.6

Mean Value 53.7 85.6 61.8 78.1

Table 4: Computation of the Rescaled Latent Variable Scores

Step 3: Computation of the Importance Values

A construct’s importance in terms of predicting another directly or indirectly linked (target)

construct in the structural model is derived from the total effect of the relationship between these two

constructs. The total effect is the sum of the direct and all the indirect effects in the structural model (Hair

et al., 2016). For example, to determine the total effect of Y1 on Y4 (Figure 1), we have to consider the

direct effect of the relationship between these two constructs (0.50) and the following three indirect effects

via Y2 and Y3, respectively:

- 11 -
 Y1 → Y2 → Y4 = 0.50 · 0.50 = 0.25

Y1 → Y2 → Y3 → Y4 = 0.50 · 0.25 · 0.25 = 0.03125

Y1 → Y3 → Y4 = 0.25 · 0.25 = 0.0625

Adding up the individual indirect effects yields the total indirect effect of Y1 on Y4, which is

approximately 0.34. The total effect of Y1 on Y4 is 0.84 (= 0.50 + 0.34), which expresses Y1’s importance

in predicting the target construct Y4. Since total effects represent the sum of direct and indirect effects, the
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

IPMA’s importance dimension supports the interpretation of complex models including meditators or

even multiple mediators.

The IPMA draws on unstandardized effects to facilitate a ceteris paribus interpretation of

predecessor constructs’ impact on the target construct. This interpretation of the unstandardized effects is

analogous to that of unstandardized weights in OLS regression models (Hair, Black, Babin, & Anderson,

2010). More precisely, by drawing on unstandardized effects, we can conclude that an increase in a certain

predecessor construct’s performance would increase the target construct’s performance by the size of its

unstandardized total effect. To determine the significance of the total effectsfor example, by means of

bias-corrected and accelerated (BCa) confidence intervals (Hair et al., 2016), researchers need to run the

bootstrapping routine. While a nonsignificant effect provides evidence that a total effect is zero in the

population, researchers should retain the corresponding construct in the IPMA since this outcome may

also represent a valuable finding (e.g., a company invests into the performance of a construct that has no

effect), which also can change with different data, for instance, in alternative contexts of the analysis.

Note: All effects denote unstandardized effects; significance testing uses the bootstrapping routine with 5,000

sample and no sign changes for determining the 95% BCs confidence intervals.

Table 5 summarizes all the total effects with respect to our target construct Y4. Note that Y3 does not

have an indirect effect on Y4; therefore, its total effect equals the direct effect of 0.25. At this point, after

computing the importance and performance values, all information required to draw the importance-

performance map is available.

- 12 -
 Are the Total
Predecessor
Total Indirect Effects on Y4
Construct Direct Effect on Y4 Total Effect on Y4
Effect on Y4 Significant?

Y1 0.50 0.34 0.84 Yes

Y2 0.50 0.06 0.56 Yes

Y3 0.25 - 0.25 Yes

Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

Note: All effects denote unstandardized effects; significance testing uses the bootstrapping routine with 5,000 sample

and no sign changes for determining the 95% BCs confidence intervals.

Table 5: Direct, Indirect and Total Effects in the IPMA

Finally, the IPMA also supports path models with moderators. However, if one path relationship in

a total effect is moderated, the interpretation of the total effect changes. More precisely, the path

coefficient estimate of a moderated effect expresses the strength of the relationship between the two

constructs when the moderator variable has the value 0 in case researchers follow standard procedure and

standardize the moderator’s indicators prior to the analysis (for more details see Hair et al., 2016). This

interpretation, however, complicates any comparison of total effects that include moderating effects with

those that lack a moderating effect. With multiple moderators in a total effect or moderated mediation

effect, the interpretation of IPMA’s importance dimension becomes difficult. Therefore, we generally

advise against the inclusion of moderators in an IPMA.

Step 4: Importance-Performance Map Creation

The IPMA focuses on one key target construct of interest in the PLS path model. Therefore, the first

step in creating an importance-performance map requires selecting the target construct of interest. In our

example, Y4 represents such a key target construct (Figure 1). The importance and performance values of

Y4’s predecessor constructs (i.e., Y1, Y2, and Y3) allow creating the importance-performance map of Y4.

- 13 -
 Table 6 summarizes the values of this map’s importance and performance dimensions—as obtained by the

previous IMPA steps.

Importance Performance

Y1 0.84 53.7

Y2 0.56 85.6
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

Y3 0.25 61.8

Mean Value 0.55 67.0

Table 6: Data of the Importance-Performance Map for Construct Y4

Scatter plotting the information shown in Table 6 allows us to create an importance-performance

map as shown in Figure 2 at the beginning of this paper. The x-axis represents the importance of Y1, Y2,

and Y3 for explaining the target construct Y4, while the y-axis depicts the performance of Y1, Y2, and Y3 in

terms of their average rescaled latent variable scores. For a better orientation, researchers may also draw

two additional lines in the importance-performance map: the mean importance value (i.e., a vertical line)

and the mean performance value (i.e., a horizontal line) of the displayed constructs (Figure 4). With regard

to our example, Y1, Y2, and Y3 have a mean importance of 0.55 and a mean performance of 67.0 (Table 6).

These two additional lines divide the importance-performance map into four areas with importance and

performance values below and above the average. Generally, when analyzing the importance-performance

map, constructs in the lower right area (i.e., above average importance and below average performance)

are of highest interest to achieve improvement, followed by the higher right, lower left and, finally, the

higher left areas. Thereby, the importance-performance map provides guidance for the prioritization of

managerial activities of high importance for the aspect underlying the selected target, but which require

performance improvements.

- 14 -
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

Figure 4: Adjusted Importance-Performance Map of Target Construct Y4

In our example, the importance-performance map (Figure 4) shows that Y1 has a relatively low

performance of 53.7. In comparison with the other constructs, Y1’s performance is slightly below average.

On the other hand, with a total effect of 0.84, this construct’s importance is particularly high. Therefore, a

one-unit increase in Y1’s performance from 53.7 to 54.7 would increase the performance of Y4 by 0.84

points from 78.10 to 78.85. Hence, when managers aim at increasing the performance of the target

construct Y4, their first priority should be to improve the performance of aspects captured by Y1, as this

construct has the highest (above average) importance, but a relatively low (below average) performance.

Aspects related to constructs Y2 and Y3 follow as a second and third priority.

Step 5: Extension of the IPMA on the Indicator Level

The IPMA is not limited to the construct level. We can also conduct an IPMA on the indicator level

to identify relevant and even more specific areas of improvement. More precisely, we can interpret the

rescaled outer weightsas reported in formative measurement modelsas an indicator’s relative

importance compared to that of the other indicators in a specific measurement model. Alternatively, the

- 15 -
 interpretation of the indicators’ relative contribution can also draw on reflective measurement models but

use the outer weights instead of the outer loadings. While the outer weights play no role in the assessment

of the reflective measurement model’s reliability and validity, they still represent each indicator’s

contribution to forming the composite variable that represents the construct in the PLS path model.

The importance values are derived from the indicators’ total effects on the target construct, which is

the result of multiplying the rescaled outer weights of a predecessor construct’s indicators with its
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

unstandardized total effect on the target construct. For example, with regard to the indicators of Y1, we

would multiply the rescaled outer weights of x11 to x14 (i.e., 0.139, 0.189, 0.215, 0.457; Table 3) with the

unstandardized total effect of Y1 on Y4 in the structural model (i.e., 0.84). This analysis yields importance

values of x11 to x14 of, respectively, 0.117, 0.159, 0.181, and 0.384. The performance values are derived

from the indicators’ mean value of the rescaled data (i.e., 79, 77.5, 59.5, and 33.5; Table 2). With this data

for all indicators of Y1, Y2, and Y3, we can create an importance-performance map as shown in Figure 5. In

this figure, rectangles represent the indicators of Y1, diamonds those of Y2, and triangles those of Y3.

Figure 5: Adjusted Importance-Performance Map of Y1, Y2, and Y3’s Indicators on Target Construct Y4

- 16 -
 Derived from this example, indicator x14 should have the highest priority for improvement, since it

has the highest relative importance, but the lowest performance. A one-unit point increase in x14’s

performance increases the performance of Y4 by x14’s importance value, which is 0.384 (ceteris paribus).

Indicators x24, x13, x23, and x12 follow with second to fifth priority. The other indicators shown in Figure 5

are less relevant for improving Y4’s performance.

Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

Empirical Example

Investigating the user acceptance and usage of new IT is a perennial theme in mainstream MIS

research. Path models explicating IT user acceptance, such as the TAM (Davis, 1989) and its extensions,

for example, the UTAUT (Venkatesh et al., 2003), are well known and have been extensively researched.

In the light of the method’s prediction orientation, researchers usually use PLS-SEM to estimate such

models. However, despite PLS-SEM’s popularity in this respect, prior research has not, to our best

knowledge, yet applied the IPMA in this context. Researchers have thus missed an opportunity to use the

available data to gain additional results and findings with which to enrich their conclusions.

In order to demonstrate the efficacy of the IPMA, we draw on data from a survey sample by Al-

Gahtani et al. (2007)1 of 722 knowledge workers in Saudi Arabia voluntarily using desktop computer

applications. These data were initially analyzed within the context of a modified UTAUT model that

synthesized model elements from various other precedent user acceptance models, such as TAM and its

extensions (e.g., TAM 2; Venkatesh & Davis, 2000). UTAUT postulates that four constructs act as

determinants of behavioral intentions to use and actual usage behavior: performance expectancy (i.e., the

degree to which individuals believe that using the system will help them attain improved job

performance), effort expectancy (i.e., the degree of ease associated with the use of the system), subjective

norm (i.e., the degree to which individuals perceive that important others believe they should use

computers), and facilitating conditions (i.e., the degree to which individuals believe that an organizational

1
The authors thank Geoffrey S. Hubona for sending and granting us permission to use the data of the study by Al-
Gahtani et al. (2007) in this article.

- 17 -
 and technical infrastructure supports the use of the system). Figure 6 shows the model and the PLS-SEM

results when using the empirical data and SmartPLS 3 software.2

Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

Figure 6: PLS Path Model and Results

The evaluation of the measurement models by means of standard evaluation criteria (e.g., Chin,

1998; Chin, 2010; Hair, Ringle, & Sarstedt, 2011, 2013; Henseler, Hubona, & Ray, 2016; Henseler,

Ringle, & Sarstedt, 2012; Roldán & Sánchez-Franco, 2012; Tenenhaus et al., 2005) supports the

measures’ reliability and validity. This also holds for discriminant validity assessment using Henseler et

al.’s (2015) recently proposed HTMT criterion, which extends the standard measures used in Al-Gahtani

et al. (2007). The results from bootstrapping with 5,000 samples using the no sign change option and the

2
The original paper also considered a further model set-up with additional moderator variables (i.e., age,
experience, gender, and voluntariness of use). Our case study draws on Al-Gahtani et al.’s (2007) initial model
without moderators. However, in light of the problems that arise in the interpretation of total effects that include
moderators, our analysis focuses on the first model in Al-Gahtani et al. (2007).

- 18 -
 95% BCa confidence intervals (Hair et al., 2016) show that all the path coefficients are statistically

significant. More specifically, performance expectancy (PE), effort expectancy (EE), and subjective norm

(SN) each have significant and positive effects on the behavioral intention to use the system (BI).

Similarly, the behavioral intention (BI) and facilitating conditions (FC) each have significant and positive

effects on the use behavior (USE). In addition, the bootstrapping results also substantiate that all total

effects on the target construct USE are significant.

Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

As a point of departure, we check the requirements for carrying out an IPMA (Step 1). After

reviewing the questionnaire, we find that the indicator data are mostly on an interval scale from 1 to 7, in

some cases from 0 to 6, and in others from 0 to 11. In respect of all the indicators, a higher value

represents a better outcome (for the description of indicators, see Tables 3 to 5 in Al-Gahtani et al., 2007).

We, therefore, do not need to reverse the scale of any of the indicators. When double clicking on the

dataset in the SmartPLS Project Explorer (Figure 7), the data view opens (Figure 8), which provides

further information on the data set (e.g., the missing value marker) and some descriptive statistics of the

indicators.

Figure 7: SmartPLS Project Explorer

- 19 -
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

Figure 8: SmartPLS Data View

Next, we inspect the signs of the outer weights. After running the PLS-SEM algorithm, SmartPLS

opens the Results Report (Figure 6), which also displays the weights of all the indicators. All the outer

weight signs are positive. Therefore, in line with IPMA Step 1, all the requirements for conducting the

analysis have been fulfilled and we can continue the analysis.

We subsequently run the IPMA by clicking on Calculate → Importance-Performance Map

Analysis (IPMA) in the menu bar. Alternatively, you can left-click on the Calculate wheel symbol in the

tool bar, and select the corresponding option in the combo box that opens. In the dialog box that opens

(Figure 9), we need to specify the target construct and decide whether to include all the predecessor

constructs of that target variable, or only those that have a direct relationship with it. We select USE as the

target construct and choose the All Predecessors of the Selected Target Construct option. Most

importantly, we need to specify each indicator’s minimum and maximum value required for the rescaling

of the data to a 0 to 100 scale. As shown in Figure 9, SmartPLS automatically reads these minimum and

- 20 -
 maximum values from the data. However, if the respondents have not made use of the full scale (e.g., the

actual minimum value is 2 instead of 1), SmartPLS cannot correctly rescale the data. Consequently, the

rescaled latent variable scores will not be between 0 and 100 but, for instance, between -5 and 95. In such

a case, we need to manually insert the true minimum value of the scale (e.g., 1 instead of 2) by clicking on

the corresponding cell in the Min column. Alternatively, we can simultaneously specify the minimum and

maximum value of all the indicators. To do so, enter the corresponding values next to Min / Max at the
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

bottom of the dialog box and click on Apply to All. In our example, all the respondents made use of the

full range of the indicator scales as indicated in the Min and Max columns of the SmartPLS Data View

(Figure 8). We therefore maintain the default settings and proceed by clicking on Start Calculation.

Figure 9: SmartPLS IPMA Dialog

SmartPLS now automatically computes the performance and importance values (Step 2 and 3) and

creates the importance-performance map (Step 4). After completing the computations, SmartPLS opens

the Results Report. Initially, it shows the results of the standardized path coefficients. Under Quality

Criteria → Importance-Performance Map [USE] (Constructs), the report includes the importance-

- 21 -
 performance map as displayed in Figure 10. Under Final Results
Total Effects, SmartPLS displays the

importance values in a matrix format. The graphical representation of the importance-performance uses

the unstandardized total effects, which you can access by clicking on the tabs Constructs,

unstandardized and Indicators unstandardized. Under Final Results
Performance / Index, you can

access the rescaled performance values of the latent and manifest variables (i.e., indicators) by means of

the tabs LV Performances and MV Performances.

Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

- 22 -
 Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

- 23 -
Figure 10: Importance-Performance Map (Construct Level)
 Not surprisingly, we find that the two direct predecessors, behavioral intention (BI) and facilitating

conditions (FC), have a particularly high importance Figure 10). Although performing at comparable

levels, the facilitating conditions (FC) construct has a considerably higher importance than the behavioral

intention (BI) construct. Managerial actions should therefore prioritize improving the performance of the

behavioral intention (BI), which can be achieved by focusing on the predecessor construct of BI with the
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

strongest impact on USE. As can be seen in Figure 10, effort expectancy (EE) has the strongest total effect

on USE.

It is important to note that the graphical representation of IMPA results differs from the graphical

PLS-SEM results illustration in SmartPLS. Instead of displaying the R² values of the endogenous latent

variables in the PLS path model (Figure 6), the IPMA results show the performance values of each latent

variable (Figure 11); instead of displaying the standardized outer loadings or weights (Figure 6), the

IPMA results show the unstandardized and rescaled outer weights of the measurement models regardless

if they are formative or reflective (Figure 11).

- 24 -
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

Figure 11: PLS Path Model and IPMA Results

To gain more specific information on how to increase the performance of constructs, the following

analyses focus on the indicator level (Step 5).3 In the results report, under Quality Criteria →

Importance-Performance Map [USE] (Indicators), SmartPLS shows the indicators’ importance-

performance map as displayed in Figure 12. For example, we find that the indicator EE2 (“It is easy for

me to become skillful using computers”) has a relatively high importance when focusing on the construct

EE, while offering some room for performance improvement. Hence, performance improvements could

focus on offering high quality computer trainings to provide users with the skills and knowledge they

need. As a direct consequence, the performance of the construct EE increases, which entails an

improvement of the construct BI and the target construct USE. Similarly, other indicators (e.g., FC1, “I

3
Note again that an IPMA on the indicator level is possible regardless of the predecessor constructs’ measurement
model specifications. However, an indicator-related analysis is particularly useful in formative measurement model
settings.

- 25 -
 have the resources and the knowledge and the ability to make use of the computer”) may gain particular

attention regarding improving the USE.

Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

- 26 -
 Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

Figure 12: Importance-Performance Map (Indicator Level)

- 27 -
 Summary and Conclusion

Review studies on the use of PLS-SEM (Hair, Sarstedt, Pieper, et al., 2012; Hair, Sarstedt, Ringle, et

al., 2012; Ringle et al., 2012; Sarstedt et al., 2014) reveal that practically all researchers rely on standard

PLS path model analysis, often ignoring more advances techniques, such as CTA-PLS (Gudergan, Ringle,

Wende, & Will, 2008), FIMIX-PLS (Hahn, Johnson, Herrmann, & Huber, 2002; Ringle, Sarstedt, &
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

Mooi, 2010; Sarstedt, Becker, Ringle, & Schwaiger, 2011; Sarstedt & Ringle, 2010), PLS-POS (Becker,

Ringle, Sarstedt, & Völckner, 2015), moderator (Henseler & Chin, 2010; Henseler & Fassott, 2010), and

multigroup analyses (Sarstedt, Henseler, et al., 2011). The IPMA belongs to this suite of often neglected

methods, but are particularly useful for generating additional findings and conclusions. By combining the

analysis of the importance and performance dimensions, the IPMA allows for prioritizing constructs to

improve a certain target construct. Expanding the analysis to the indicator level allows for identifying the

most important areas of specific actions. These results are, for example, particularly important in studies

researching the differing impacts that certain construct dimensions have on a phenomenon such as

corporate reputation (e.g., Sarstedt, Wilczynski, & Melewar, 2013), or customer satisfaction (e.g., Ringle,

Sarstedt, & Zimmermann, 2011). Another extension of the IPMA’s use is in the context of a multigroup

analysis. The IPMA allows for contrasting group results and for developing specific conclusions in respect

of each group (Rigdon, Ringle, & Sarstedt, 2010; Rigdon et al., 2011; Schloderer et al., 2014). However,

researchers should refrain from using the IPMA if the analysis does not meet the requirements mentioned

in Step 1 of the IPMA procedure (Figure 3) such as having only positive outer weights estimates in the

measurement models.

As the IPMA assumes linear relationships, future research could focus on nonlinear IPMA results

(Anderson & Mittal, 2000; Eskildsen & Kristensen, 2006; Mittal, Ross, & Baldasare, 1998), making the

analysis even more useful. For example, in the context of the Kano model, an IPMA could consider the

differing role of delighters and basic needs (e.g., Kano, Seraku, Takahashi, & Tsuji, 1984). If the

performance of a delighter construct exceeds a certain threshold, further improvement of this construct’s

- 28 -
 performance improves the target construct exponentially. Conversely, performance decreases in this

construct generally have a lower effect on the target construct. The opposite holds for basic needs, where

decreases in a corresponding construct’s performance result in steep decreases in the target construct’s

performance. Exceeding the performance of basic needs above a certain threshold will, however, only

marginally increase the target construct’s performance. In this context, the penalty-reward contrast

analysis of IPMA results (Matzler & Sauerwein, 2002; Matzler, Sauerwein, & Heischmidt, 2003) could be
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

another promising avenue for future research. Given IPMA’s capabilities and the additional benefit of

potential extensions to nonlinear effects, we expect that the much broader use of the method in future

studies will extend the results presentations and allow more elaborate findings and conclusions.

Acknowledgements

The authors thank Geoffrey S. Hubona for sending and granting us permission to use the data of
the study by Al-Gahtani et al. (2007). This article uses the statistical software SmartPLS 3
([Link] Ringle acknowledges a financial interest in SmartPLS.

References

Al-Gahtani, S. S., Hubona, G. S. and Wang, J. (2007), "Information Technology (IT) in Saudi Arabia:

Culture and the Acceptance and Use of IT", Information & Management, Vol. 44 No. 8, pp. 681-

691.

Anderson, E. W. and Fornell, C. G. (2000), "Foundations of the American Customer Satisfaction Index ",

Total Quality Management, Vol. 11 No. 7, pp. 869-882.

Anderson, E. W. and Mittal, V. (2000), "Strengthening the Satisfaction-Profit Chain", Journal of Service

Research, Vol. 3 No. 2, pp. 107-120

Becker, J.-M., Ringle, C. M., Sarstedt, M. and Völckner, F. (2015), "How Collinearity Affects Mixture

Regression Results", Marketing Letters, Vol. in press.

Bollen, K. A. and Bauldry, S. (2011), "Three Cs in Measurement Models: Causal Indicators, Composite

Indicators, and Covariates", Psychological Methods, Vol. 16 No. 3, pp. 265-284.

- 29 -
 Bollen, K. A. and Diamantopoulos, A. (in press), " In Defense of Causal–Formative Indicators: A

Minority Report", Psychological Methods.

Chin, W. W. (1998), "The Partial Least Squares Approach to Structural Equation Modeling", in

Marcoulides, G. A. (Ed.), Modern Methods for Business Research. Erlbaum, Mahwah, pp. 295-358.

Chin, W. W. (2010), "How to Write Up and Report PLS Analyses", in Esposito Vinzi, V., Chin, W. W.,

Henseler, J. and Wang, H. (Eds.), Handbook of Partial Least Squares: Concepts, Methods and
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

Applications (Springer Handbooks of Computational Statistics Series, vol. II). Springer,

Heidelberg, Dordrecht, London, New York, pp. 655-690.

Davis, F. D. (1989), "Perceived Usefulness, Perceived Ease of Use, and User Acceptance of Information

Technology", MIS Quarterly, Vol. 13 No. 3, pp. 319-340.

Davis, F. D., Bagozzi, R. P. and Warshaw, P. R. (1989), "User Acceptance of Computer Technology: A

Comparison of Two Theoretical Models", Management Science, Vol. 35 No. 8, pp. 982-1003.

Dijkstra, T. K. and Henseler, J. (2015), "Consistent Partial Least Squares Path Modeling", MIS Quarterly,

Vol. 39 No. 2, pp. 297-316.

do Valle, P. O. and Assaker, G. (2015), "Using Partial Least Squares Structural Equation Modeling in

Tourism Research: A Review of Past Research and Recommendations for Future Applications",

Journal of Travel Research, Vol. in press.

Eskildsen, J. K. and Kristensen, K. (2006), "Enhancing Importance‐Performance Analysis", International

Journal of Productivity and Performance Management, Vol. 55 No. 1, pp. 40-60.

Fornell, C. G., Johnson, M. D., Anderson, E. W., Cha, J. and Bryant, B. E. (1996), "The American

Customer Satisfaction Index: Nature, Purpose, and Findings", Journal of Marketing, Vol. 60 No. 4,

pp. 7-18.

Garson, G. D. (2014), Partial Least Squares Regression and Structural Equation Models, Statistical

Associates, Asheboro.

- 30 -
 Grønholdt, L., Martensen, A., Jørgensen, S. and Jensen, M. J. (2015), "Customer Experience Management

and Business Performance", International Journal of Quality and Service Sciences, Vol. 7 No. 1,

pp. 90-106.

Gudergan, S. P., Ringle, C. M., Wende, S. and Will, A. (2008), "Confirmatory Tetrad Analysis in PLS

Path Modeling", Journal of Business Research, Vol. 61 No. 12, pp. 1238-1249.

Hahn, C., Johnson, M. D., Herrmann, A. and Huber, F. (2002), "Capturing Customer Heterogeneity Using
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

a Finite Mixture PLS Approach", Schmalenbach Business Review, Vol. 54 No. 3, pp. 243-269.

Hair, J. F., Black, W. C., Babin, B. J. and Anderson, R. E. (2010), Multivariate Data Analysis, Prentice

Hall, Englewood Cliffs.

Hair, J. F., Hult, G. T. M., Ringle, C. M. and Sarstedt, M. (2016), A Primer on Partial Least Squares

Structural Equation Modeling (PLS-SEM), Sage, Thousand Oaks, CA.

Hair, J. F., Ringle, C. M. and Sarstedt, M. (2011), "PLS-SEM: Indeed a Silver Bullet", Journal of

Marketing Theory and Practice, Vol. 19 No. 2, pp. 139-151.

Hair, J. F., Ringle, C. M. and Sarstedt, M. (2013), "Partial Least Squares Structural Equation Modeling:

Rigorous Applications, Better Results and Higher Acceptance", Long Range Planning, Vol. 46 No.

1-2, pp. 1-12.

Hair, J. F., Sarstedt, M., Pieper, T. M. and Ringle, C. M. (2012), "The Use of Partial Least Squares

Structural Equation Modeling in Strategic Management Research: A Review of Past Practices and

Recommendations for Future Applications", Long Range Planning, Vol. 45 No. 5-6, pp. 320-340.

Hair, J. F., Sarstedt, M., Ringle, C. M. and Mena, J. A. (2012), "An Assessment of the Use of Partial Least

Squares Structural Equation Modeling in Marketing Research", Journal of the Academy of

Marketing Science, Vol. 40 No. 3, pp. 414-433.

Henseler, J. and Chin, W. W. (2010), "A Comparison of Approaches for the Analysis of Interaction

Effects Between Latent Variables Using Partial Least Squares Path Modeling", Structural Equation

Modeling, Vol. 17 No. 1, pp. 82-109.

- 31 -
 Henseler, J. and Fassott, G. (2010), "Testing Moderating Effects in PLS Path Models: An Illustration of

Available Procedures", in Esposito Vinzi, V., Chin, W. W., Henseler, J. and Wang, H. (Eds.),

Handbook of Partial Least Squares: Concepts, Methods and Applications (Springer Handbooks of

Computational Statistics Series, vol. II). Springer, Heidelberg, Dordrecht, London, New York, pp.

713-735.

Henseler, J., Hubona, G. S. and Ray, P. A. (2016), "Using PLS Path Modeling in New Technology
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

Research: Updated Guidelines", Industrial Management & Data Systems, Vol. 116 No. 1, pp. in

press.

Henseler, J., Ringle, C. M. and Sarstedt, M. (2012), "Using Partial Least Squares Path Modeling in

International Advertising Research: Basic Concepts and Recent Issues", in Okazaki, S. (Ed.),

Handbook of Research in International Advertising. Edward Elgar Publishing, Cheltenham, pp.

252-276.

Höck, C., Ringle, C. M. and Sarstedt, M. (2010), "Management of Multi-Purpose Stadiums: Importance

and Performance Measurement of Service Interfaces", International Journal of Services Technology

and Management, Vol. 14 No. 2/3, pp. 188-207.

Kano, N., Seraku, N., Takahashi, F. and Tsuji, S.-i. (1984), "Attractive Quality and Must-be Quality",

Journal of the Japanese Society for Quality Control, Vol. 14 No. 2, pp. 147-156.

Kristensen, K., Martensen, A. and Grønholdt, L. (2000), "Customer Satisfaction Measurement at Post

Denmark: Results of Application of the European Customer Satisfaction Index Methodology",

Total Quality Management, Vol. 11 No. 7, pp. 1007-1015.

Lee, L., Petter, S., Fayard, D. and Robinson, S. (2011), "On the Use of Partial Least Squares Path

Modeling in Accounting Research", International Journal of Accounting Information Systems, Vol.

12 No. 4, pp. 305-328.

Lohmöller, J.-B. (1989), Latent Variable Path Modeling with Partial Least Squares, Physica, Heidelberg.

- 32 -
 Martensen, A., Dahlgaard, J. J., Park‐Dahlgaard, S. M. and Grønholdt, L. (2007), "Measuring and

Diagnosing Innovation Excellence - Simple Contra Advanced Approaches: A Danish Study",

Measuring Business Excellence, Vol. 11 No. 4, pp. 51-65.

Martensen, A. and Grønholdt, L. (2010), "Measuring and Managing Brand Equity: A Study with Focus on

Product and Service Quality in Banking", International Journal of Quality and Service Sciences,

Vol. 2 No. 3, pp. 300-316.

Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

Martilla, J. A. and James, J. C. (1977), "Importance-Performance Analysis", Journal of Marketing, Vol.

41 No. 1, pp. 77-79.

Matzler, K. and Sauerwein, E. (2002), "The Factor Structure of Customer Satisfaction: An Empirical Test

of the Importance Grid and the Penalty‐Reward‐Contrast Analysis", International Journal of

Service Industry Management, Vol. 13 No. 4, pp. 314-332.

Matzler, K., Sauerwein, E. and Heischmidt, K. A. (2003), "Importance-Performance Analysis Revisited:

The Role of the Factor Structure of Customer Satisfaction", Service Industries Journal, Vol. 23 No.

2, pp. 112-129.

Mittal, V., Ross, W. T., Jr. and Baldasare, P. M. (1998), "The Asymmetric Impact of Negative and

Positive Attribute-Level Performance on Overall Satisfaction and Repurchase Intentions", Journal

of Marketing, Vol. 62 No. 1, pp. 33-47.

Peng, D. X. and Lai, F. (2012), "Using Partial Least Squares in Operations Management Research: A

Practical Guideline and Summary of Past Research", Journal of Operations Management, Vol. 30

No. 6, pp. 467–480.

Richter, N. F., Sinkovics, R. R., Ringle, C. M. and Schlägel, C. (2015), "A Critical Look at the Use of

SEM in International Business Research ", International Marketing Review, Vol. forthcoming.

Rigdon, E. E. (2012), "Rethinking Partial Least Squares Path Modeling: In Praise of Simple Methods",

Long Range Planning, Vol. 45 No. 5-6, pp. 341-358.

- 33 -
 Rigdon, E. E. (2013), "Partial Least Squares Path Modeling", in Hancock, G. R. and Mueller, R. O. (Eds.),

Structural Equation Modeling. A Second Course. Information Age Publishing, Charlotte NC, pp.

81-116.

Rigdon, E. E. (2014), "Rethinking Partial Least Squares Path Modeling: Breaking Chains and Forging

Ahead", Long Range Planning, Vol. 47 No. 3, pp. 161-167.

Rigdon, E. E., Ringle, C. M. and Sarstedt, M. (2010), "Structural Modeling of Heterogeneous Data with
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

Partial Least Squares", in Malhotra, N. K. (Ed.), Review of Marketing Research. Sharpe, Armonk,

pp. 255-296.

Rigdon, E. E., Ringle, C. M., Sarstedt, M. and Gudergan, S. P. (2011), "Assessing Heterogeneity in

Customer Satisfaction Studies: Across Industry Similarities and Within Industry Differences",

Advances in International Marketing, Vol. 22, pp. 169-194.

Ringle, C. M., Sarstedt, M. and Mooi, E. A. (2010), "Response-Based Segmentation Using Finite Mixture

Partial Least Squares: Theoretical Foundations and an Application to American Customer

Satisfaction Index Data", Annals of Information Systems, Vol. 8, pp. 19-49.

Ringle, C. M., Sarstedt, M. and Straub, D. W. (2012), "A Critical Look at the Use of PLS-SEM in MIS

Quarterly", MIS Quarterly, Vol. 36 No. 1, pp. iii-xiv.

Ringle, C. M., Sarstedt, M. and Zimmermann, L. (2011), "Customer Satisfaction with Commercial

Airlines: The Role of Perceived Safety and Purpose of Travel", Journal of Marketing Theory and

Practice, Vol. 19 No. 4, pp. 459-472.

Ringle, C. M., Wende, S. and Becker, J.-M. (2015), "SmartPLS 3", in. SmartPLS GmbH, Bönningstedt.

Roldán, J. L. and Sánchez-Franco, M. J. (2012), "Variance-Based Structural Equation Modeling:

Guidelines for Using Partial Least Squares in Information Systems Research", in Mora, M.,

Gelman, O., Steenkamp, A. L. and Raisinghani, M. (Eds.), Research Methodologies, Innovations

and Philosophies in Software Systems Engineering and Information Systems. IGI Global, Hershey,

PA, pp. 193-221.

- 34 -
 Sarstedt, M., Becker, J.-M., Ringle, C. M. and Schwaiger, M. (2011), "Uncovering and Treating

Unobserved Heterogeneity with FIMIX-PLS: Which Model Selection Criterion Provides an

Appropriate Number of Segments?", Schmalenbach Business Review, Vol. 63 No. 1, pp. 34-62.

Sarstedt, M., Henseler, J. and Ringle, C. M. (2011), "Multi-Group Analysis in Partial Least Squares (PLS)

Path Modeling: Alternative Methods and Empirical Results", in Sarstedt, M., Schwaiger, M. and

Taylor, C. R. (Eds.), Advances in International Marketing, Volume 22. Emerald, Bingley, pp. 195-
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

218.

Sarstedt, M. and Mooi, E. A. (2014), A Concise Guide to Market Research: The Process, Data, and

Methods Using IBM SPSS Statistics, Springer, Berlin et al.

Sarstedt, M. and Ringle, C. M. (2010), "Treating Unobserved Heterogeneity in PLS Path Modelling: A

Comparison of FIMIX-PLS with Different Data Analysis Strategies", Journal of Applied Statistics,

Vol. 37 No. 8, pp. 1299-1318.

Sarstedt, M., Ringle, C. M., Smith, D., Reams, R. and Hair, J. F. (2014), "Partial Least Squares Structural

Equation Modeling (PLS-SEM): A Useful Tool for Family Business Researchers", Journal of

Family Business Strategy, Vol. 5 No. 1, pp. 105-115.

Sarstedt, M., Wilczynski, P. and Melewar, T. C. (2013), "Measuring Reputation in Global Markets - A

Comparison of Reputation Measures' Convergent and Criterion Validities ", Journal of World

Business, Vol. 48 No. 3, pp. 329-339.

Schloderer, M. P., Sarstedt, M. and Ringle, C. M. (2014), "The Relevance of Reputation in the Nonprofit

Sector: The Moderating Effect of Socio-Demographic Characteristics", International Journal of

Nonprofit and Voluntary Sector Marketing, Vol. 19 No. 2, pp. 110-126.

Slack, N. (1994), "The Importance-Performance Matrix as a Determinant of Improvement Priority",

International Journal of Operations and Production Management, Vol. 44 No. 5, pp. 59-75.

Tenenhaus, M., Esposito Vinzi, V., Chatelin, Y.-M. and Lauro, C. (2005), "PLS Path Modeling",

Computational Statistics & Data Analysis, Vol. 48 No. 1, pp. 159-205.

- 35 -
 Venkatesh, V. and Davis, F. D. (2000), "A Theoretical Extension of the Technology Acceptance Model:

Four Longitudinal Field Studies", Management Science, Vol. 46 No. 186-204, pp. 2.

Venkatesh, V., Morris, M. G., Davis, G. B. and Davis, F. D. (2003), "User Acceptance of Information

Technology: Toward a Unified View", MIS Quarterly, Vol. 27 No. 3, pp. 425-478.

Völckner, F., Sattler, H., Hennig-Thurau, T. and Ringle, C. M. (2010), "The Role of Parent Brand Quality

for Service Brand Extension Success", Journal of Service Research, Vol. 13 No. 4, pp. 359-361.
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

Wold, H. O. A. (1982), "Soft Modeling: The Basic Design and Some Extensions", in Jöreskog, K. G. and

Wold, H. O. A. (Eds.), Systems Under Indirect Observations: Part II. North-Holland, Amsterdam,

pp. 1-54.

Author Biographies

Christian M. Ringle is a Chaired Professor of Management at the Hamburg University of

Technology (TUHH) and Conjoint Professor at the Faculty of Business Law at the University of

Newcastle. His widely published research addresses the management of organizations, strategic and

human resource management, marketing, and quantitative methods for business and market research. He

is cofounder of SmartPLS ([Link] a statistical software tool with a graphical user

interface for partial least squares structural equation modeling (PLS-SEM). He regularly teaches doctoral

seminars on multivariate statistics, the PLS-SEM method, and the use of SmartPLS worldwide. More

information about Christian Ringle can be found at [Link]

[Link].

Marko Sarstedt is Chaired Professor of Marketing at the Otto-von-Guericke-University

Magdeburg and Conjoint Professor to the Faculty of Business and Law at the University of Newcastle. His

main research is in the application and advancement of structural equation modeling methods to improve

the understanding of consumer behavior and to improve marketing decision making. His research has been

published in journals such as Journal of Marketing Research, Journal of the Academy of Marketing

Science, Organizational Research Methods, MIS Quarterly, International Journal of Research in

- 36 -
 Marketing, Long Range Planning. He regularly teaches doctoral seminars on multivariate statistics,

structural equation modeling, and measurement worldwide.

Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

- 37 -
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

277x176mm (72 x 72 DPI)

Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)
Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

159x137mm (72 x 72 DPI)

Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

319x109mm (72 x 72 DPI)

Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

277x176mm (72 x 72 DPI)

Downloaded by Ecole Polytechnique Federale de Lausanne At 02:41 22 September 2016 (PT)

319x109mm (72 x 72 DPI)