Handbook of Partial Least Squares

In this chapter it is shown that the PLS-algorithms typically converge if the covariance matrix of the indicators satisfies (approximately) the “basic design”, a factor analysis type of model. The algorithms produce solutions to fixed point equations; the solutions are smooth functions of the sample covariance matrix of the indicators. If the latter matrix is asymptotically normal, the PLS-estimators will share this property. The probability limits, under the basic design, of the PLS-estimators for loadings, correlations, multiple R’s, coefficients of structural equations et cetera will differ from the true values. But the difference is decreasing, tending to zero, in the “quality” of the PLS estimators for the latent variables. It is indicated how to correct for the discrepancy between true values and the probability limits. We deemphasize the “normality”-issue in discussions about PLS versus ML: in employing either method one is not required to subscribe to normality; they are “just” different ways of extracting information from second-order moments.

We also propose a new “back-to-basics” research program, moving away from factor analysis models and returning to the original object of constructing indices that extract information from high-dimensional data in a predictive, useful way. For the generic case we would construct informative linear compounds, whose constituent indicators have non-negative weights as well as non-negative loadings, satisfying constraints implied by the path diagram. Cross-validation could settle the choice between various competing specifications. In short: we argue for an upgrade of principal components and canonical variables analysis.

In this chapter the authors first present the basic algorithm of PLS Path Modeling by discussing some recently proposed estimation options. Namely, they introduce the development of new estimation modes and schemes for multidimensional (formative) constructs, i.e. the use of PLS Regression for formative indicators, and the use of path analysis on latent variable scores to estimate path coefficients Furthermore, they focus on the quality indexes classically used to assess the performance of the model in terms of explained variances. They also present some recent developments in PLS Path Modeling framework for model assessment and improvement, including a non-parametric GoF-based procedure for assessing the statistical significance of path coefficients. Finally, they discuss the REBUS-PLS algorithm that enables to improve the prediction performance of the model by capturing unobserved heterogeneity. The chapter ends with a brief sketch of open issues in the area that, in the Authors’ opinion, currently represent major research challenges.

The goal of PLS path modeling is primarily to estimate the variance of endogenous constructs and in turn their respective manifest variables (if reflective). Models with significant jackknife or bootstrap parameter estimates may still be considered invalid in a predictive sense. In this chapter, the objective is to shift from that of assessing the significance of parameter estimates (e.g., loadings and structural paths) to that of predictive validity. Specifically, this chapter examines how predictive indicator weights estimated for a particular PLS structural model are when applied on new data from the same population. Bootstrap resampling is used to create new data sets where new R-square measures are obtained for each endogenous construct in a model. The weighted summed (WSD) R-square represents how well the original sample weights predict when given new data (i.e., a new bootstrap sample). In contrast, the simple summed (SSD) R-square examines the predictiveness using the simpler approach of unit weights. Such an approach is equivalent to performing a traditional path analysis using simple summed scale scores. A relative performance index (RPI) based on the WSD and SSD estimates is created to represent the degree to which the PLS weights yield better predictiveness for endogenous constructs than the simpler procedure of performing regression after simple summing of indicators. In addition, a Performance from Optimized Summed Index (PFO) is obtained by contrasting the WSD R-squares to the R-squares obtained when the PLS algorithm is used on each new bootstrap data set. Results from two studies are presented. In the first study, 14 data sets of sample size 1,000 were created to represent two different structural models (i.e., medium versus high R-square) consisting of one endogenous and three exogenous constructs across seven different measurement scenarios (e.g., parallel versus heterogenous loadings). Five-hundred bootstrap cross validation data sets were generated for each of 14 data sets. In study 2, simulated data based on the population model conforming to the same scenarios in study 1 were used instead of the bootstrap samples in part to examine the accuracy of the bootstrapping approach. Overall, in contrast to Q-square which examines predictive relevance at the indicator level, the RPI and PFO indices are shown to provide additional information to assess predictive relevance of PLS estimates at the construct level. Moreover, it is argued that this approach can be applied to other same set data indices such as AVE (Fornell C, Larcker D, J Mark Res 18:39–50, 1981) and GoF (Tenenhaus M, Amato S, Esposito Vinzi V, Proceedings of the XLII SIS (Italian Statistical Society) Scientific Meeting, vol. Contributed Papers, 739–742, CLEUP, Padova, Italy, 2004) to yield RPI-AVE, PFO-AVE. RPI-GoF, and PFO-GoF indices.

A situation where

J

blocks of variables

X

1

,

…

,

X

J

are observed on the same set of individuals is considered in this paper. A factor analysis approach is applied to blocks instead of variables. The latent variables (LV’s) of each block should well explain their own block and at the same time the latent variables of same order should be as highly correlated as possible (positively or in absolute value). Two path models can be used in order to obtain the first order latent variables. The first one is related to confirmatory factor analysis: each LV related to one block is connected to all the LV’s related to the other blocks. Then, PLS path modeling is used with mode A and centroid scheme. Use of mode B with centroid and factorial schemes is also discussed. The second model is related to hierarchical factor analysis. A causal model is built by relating the LV’s of each block

X

j

to the LV of the super-block

X

J

+ 1

obtained by concatenation of

X

1

,

…

,

X

J

. Using PLS estimation of this model with mode A and path-weighting scheme gives an adequate solution for finding the first order latent variables. The use of mode B with centroid and factorial schemes is also discussed. The higher order latent variables are found by using the same algorithms on the deflated blocks. The first approach is compared with the MAXDIFF/MAXBET Van de Geer’s algorithm (1984) and the second one with the ACOM algorithm (Chessel and Hanafi, 1996). Sensory data describing Loire wines are used to illustrate these methods.

The objective of this paper is to describe the use of unweighted least squares (ULS) structural equation modeling (SEM) and partial least squares (PLS) path modeling in a regression model relating two blocks of binary variables, when a group effect can influence the relationship. Two sets of binary variables are available. The first set is defined by one block

X

of predictors and the second set by one block

Y

of responses. PLS regression could be used to relate the responses

Y

to the predictors

X

, taking into account the block structure. However, for multigroup data, this model cannot be used because the path coefficients can be different from one group to another. The relationship between

Y

and

X

is studied in the context of structural equation modeling. A group effect

A

can affect the measurement model (relating the manifest variables (MVs) to their latent variables (LVs)) and the structural equation model (relating the

Y

-LV to the

X

-LV). In this paper, we wish to study the impact of the group effect on the structural model only, supposing that there is no group effect on the measurement model. This approach has the main advantage of allowing a description of the group effect (main and interaction effects) at the LV level instead of the MV level. Then, an application of this methodology on the data of a questionnaire investigating sun exposure behavior is presented.

This study presents a new algorithm for estimating causal models based on multiblock PLS method. This new algorithm is tested in a particular post-consumption behavior with the aim of validating a complex system of relations between antecedents of value, perceived value, satisfaction and loyalty. The results are compared with the classical LVPLS method: both methods support the proposed structural relations, but the explained variance is slightly higher with the new algorithm.

To date, multi-group comparison of Partial Least Square (PLS) models where differences in path estimates for different sampled populations have been relatively naive. Often, researchers simply examine and discuss the difference in magnitude of specific model path estimates from two or more data sets. When evaluating the significance of path differences, a

t

-test based on the pooled standard errors obtained via a resampling procedure such as bootstrapping from each data set is made. Yet problems can occur if the assumption of normal population or similar sample size is made. This paper provides an introduction to an alternative distribution free approach based on an approximate randomization test – where a subset of all possible data permutations between sample groups is made. The performance of this permutation procedure is tested on both simulated data and a study exploring the differences of factors that impact outsourcing between the countries of US and Germany. Furthermore, as an initial examination of the consistency of this new procedure, the outsourcing results are compared with those obtained from using covariance based SEM (AMOS 7).

In wide range of applications for empirical data analysis, the assumption that data is collected from a single homogeneous population is often unrealistic. In particular, the identification of different groups of consumers and their appropriate consideration in partial least squares (PLS) path modeling constitutes a critical issue in marketing. In this work, we introduce a finite mixture PLS software implementation which separates data on the basis of the estimates’ heterogeneity in the inner path model. Numerical examples using experimental as well as empirical data allow the verification of the methodology’s effectiveness and usefulness. The approach permits a reliable identification of distinctive customer segments along with characteristic estimates for relationships between latent variables. Researchers and practitioners can employ this method as a model evaluation technique and thereby assure that results on the aggregate data level are not affected by unobserved heterogeneity in the inner path model estimates. Otherwise, the analysis provides further indications on how to treat that problem by forming groups of data in order to perform a multi-group path analysis.

Structural Equation Modelling methods traditionally assume the homogeneity of all the units on which a model is estimated. In many cases, however, this assumption may turn to be false; the presence of latent classes not accounted for by the global model may lead to biased or erroneous results in terms of model parameters and model quality. The traditional multi-group approach to classification is often unsatisfying for several reasons; above all because it leads to classes homogeneous only with respect to external criteria and not to the theoretical model itself.

In this paper, a prediction-oriented classification method in PLS Path Modelling is proposed. Following PLS Typological Regression, the proposed methodology aims at identifying classes of units showing the lowest distance from the models in the space of the dependent variables, according to PLS predictive oriented logic. Hence, the obtained groups are homogeneous with respect to the defined path model. An application to real data in the study of customers’ satisfaction and loyalty will be shown.

In PLS approach, it is frequently assumed that the blocks of variables satisfy the assumption of unidimensionality. In order to fulfill at best this hypothesis, we use clustering methods of variables. We illustrate the conjoint use of variables clustering and PLS structural equations modeling on data provided by PSA Company (Peugeot Citroën) on customers’ satisfaction. The data are satisfaction scores on 32 manifest variables given by 2,922 customers.

In this chapter we focus on design of PLS structural equation modeling with respect to satisfaction studies in general. Previous studies have found the PLS technique to be affected by things as the skewness of manifest variables, multicollinearity between latent variables, misspecification, question order, sample size as well as the size of the path coefficients (Cassel et al.

1999

; Auh et al.

2003

; Eskildsen and Kristensen 2005; Kristensen and Eskildsen

2005a,b

). In this chapter we expand on these contributions in order to provide the reader with recommendations on all aspects included in designing PLS-based satisfaction studies.

The recommendations are based on an empirical PLS project conducted at the Aarhus School of Business, Center for Corporate Performance. Within this project five different studies have been conducted that cover a variety of aspects of designing PLS-based satisfaction studies.The data used in subsequent sections comes from a variety of sources. In relation to the empirical PLS project at the Aarhus School off Business the following five different studies have been conducted:

Scale study

Empirical experiment

Simulation study – data collection

Simulation study – missing values

Empirical study of model specification for a customer satisfaction model

Bootstrap is a resampling technique proposed by Efron (The Annals of Statistics 7:1–26, 1979). It has been used in many fields, but in case of missing data studies one can find only a few references.

Most studies in marketing research are based on questionnaires, that, for several reasons present missing responses. The missing data problem is a common issue in market research. Here, a customer satisfaction model following the ACSI barometer from Fornell (Journal of Marketing 60(4):7–18, 1996; The American customer satisfaction index: methodology report. Michigan: University of Michigan Business School, 1998) will be considered. Sometimes not all customers experience all services or products. Therefore, we may have to deal with missing data, taking the risk of reaching non-significant impacts of these drivers on Customer Satisfaction and resulting in inaccurate inferences. To estimate the main drivers of Customer Satisfaction, Structural Equation Models methodology is applied (Peters and Enders, Journal of Targeting Measurement and Analysis for Marketing 11(1):81–95, 2002).

For a case study in mobile telecommunications several missing data imputation techniques were reviewed and used to complete the data set. Bootstrap methodology was also considered jointly with imputation techniques to complete the data set. Finally, using Partial Least Squares (PLS) algorithm we could compare the above procedures. It suggests that bootstrapping before imputation can be a promising idea.

Although PLS is a well established tool to estimate structural equation models, more work is still needed in order to better understand its relative merits when compared to likelihood methods. This paper aims to contribute to a better understanding of PLS and likelihood estimators’ properties, through the comparison and evaluation of these estimation methods for structural equation models based on customer satisfaction data. A Monte Carlo simulation is used to compare the two estimation methods. The model used in the simulation is the ECSI (European Customer Satisfaction Index) model, constituted by 6 latent variables (image, expectations, perceived quality, perceived value, customer satisfaction and customer loyalty). The simulation is conducted in the context of symmetric and skewed response data and formative blocks, which constitute the typical framework of customer satisfaction measurement. In the simulation we analyze the ability of each method to adequately estimate the inner model coefficients and the indicator loadings. The estimators are analyzed both in terms of bias and precision. Results have shown that globally PLS estimates are generally better than covariance-based estimates both in terms of bias and precision. This is particularly true when estimating the model with skewed response data or a formative block, since for the model based on symmetric data the two methods have shown a similar performance.

Partial least squares (PLS) estimates of structural equation model path coefficients are believed to produce more accurate estimates than those obtained with covariance structure analysis (CVA) using maximum likelihood estimation (MLE) when one or more of the MLE assumptions are not met. However, there exists no empirical support for this belief or for the specific conditions under which it will occur. MLE-based CVA will also break down or produce improper solutions whereas PLS will not. This study uses simulated data to estimate parameters for a model with five independent latent variables and one dependent latent variable under various assumption conditions. Data from customer satisfaction studies were used to identify the form of typical field-based survey distributions. Our results show that PLS produces more accurate path coefficients estimates when sample sizes are less than 500, independent latent variables are correlated, and measures per latent variable are less than 4. Method accuracy does not vary when the MLE multinormal distribution assumption is violated or when the data do not fit the theoretical structure very well. Both procedures are more accurate when the independent variables are uncorrelated, but MLE estimations break down more frequently under this condition, especially when combined with sample sizes of less than 100 and only two measures per latent variable.

Data mining by means of projection methods such as PLS (projection to latent structures), and their extensions is discussed. The most common data analytical questions in data mining are covered, and illustrated with examples.

(a)

Clustering, i.e., finding and interpreting “natural” groups in the data

(b)

Classification and identification, e.g., biologically active compounds vs inactive

(c)

Quantitative relationships between different sets of variables, e.g., finding variables related to quality of a product, or related to time, seasonal or/and geographical change

Sub-problems occurring in both (a) to (c) are discussed.

(1)

Identification of outliers and their aberrant data profiles

(2)

Finding the dominating variables and their joint relationships

(3)

Making predictions for new samples

The use of graphics for the contextual interpretation of results is emphasized.

With many variables and few observations (samples) – a common situation in data mining – the risk to obtain spurious models is substantial. Spurious models look great for the training set data, but give miserable predictions for new samples. Hence, the validation of the data analytical results is essential, and approaches for that are discussed.

In consumer science it is common to study how various products are liked or ranked by various consumers. In this context, it is important to check if there are different consumer groups with different product preference patterns. If systematic consumer grouping is detected, it is important to determine the person characteristics which differentiate between these consumer segments, so that they can be reached selectively. Likewise it is important to determine the product characteristics that consumer segments seem to respond differently to.

Consumer preference data are usually rather noisy. The products ×persons data table (\vec{

X

}

1

) usually produced in consumer preference studies may therefore be supplemented with two types of background information: a products ×product-property data table (\vec{

X

}

2

) and a person ×person-property data table (\vec{

X

}

3

). These additional data may be used for stabilizing the data modeling of the preference data \vec{

X

}

1

statistically. Moreover, they can reveal the product-properties that are responded to differently by the different consumer segments, and the person-properties that characterize these different segments. The present chapter outlines a recent approach to analyzing the three types of data tables in an integrated fashion and presents new modeling methods in this context.

In data analysis of social, economic and technical fields, compositional data is widely used in problems of proportions to the whole. This paper develops regression modelling methods of compositional data, discussing the relationships of one compositional data to one or more than one compositional data and the interrelationship of multiple compositional data. By combining centered logratio transformation proposed by Aitchison (The Statistical Analysis of Compositional Data, Chapman and Hall, 1986) with Partial Least Squares (PLS) related techniques, that is PLS regression, hierarchical PLS and PLS path modelling, respectively, particular difficulties in compositional data regression modelling such as sum to unit constraint, high multicollinearity of the transformed compositional data and hierarchical relationships of multiple compositional data, are all successfully resolved; moreover, the modelling results rightly satisfies the theoretical requirement of logcontrast. Accordingly, case studies of employment structure analysis of Beijing’s three industries also illustrate high goodness-of-fit and powerful explainability of the models.

While in consumer research the “Cronbach’s α – LISREL”-paradigm has emerged for a better separation of measurement errors and structural relationships, it is shown here that studies involving an evaluation of the effectiveness of marketing or organizational strategies based on structural relationships require the application of PLS. This is because we no longer distinguish between constructs and their reflecting measures but rather between abstract marketing policies (constructs) and their forming detailed marketing instruments (indicators). It is shown with the help of examples from literature that many studies of this type applying LISREL have been misspecified and would have better made use of the PLS approach. I also demonstrate the appropriate use of PLS in a study of success factors for e-businesses. I conclude with recommendations on the appropriate design of success factor studies, including the use of higher-order constructs and the validation of such studies.

Structural equation modelling (SEM) has been increasingly utilized in marketing and management areas. This increasing deployment of SEM suggests that a comparison should be made of the different SEM approaches. This would help researchers choose the SEM approach that is most appropriate for their studies. After a brief review of the SEM theoretical background, this study analyzes two models with different sample sizes by applying two different SEM techniques to the same set of data. The two SEM techniques compared are: Covariance-based SEM (CBSEM) – specifically, maximum likelihood (ML) estimation – and Partial Least Squares (PLS). After presenting the study findings, the paper provides insights regarding when researchers should analyze models with CBSEM and when with PLS. Finally, practical suggestions concerning PLS use are presented and we discuss whether researcher considered these.

Brands play an important role in consumers’ daily life and can represent a big asset for companies owning them. Owing to the very close relationship between brands and consumers, and the specific nature of branded products as an element of consumer life style, the branded goods industry needs to extend its knowledge of the process of brand preference formation in order to enhance brand equity.

This chapter show how Partial Least Squares (PLS) modeling can be used to successfully test complex models where other approaches would fail due to the high number of relationships, constructs and indicators. Here, PLS modeling is applied to brand preference formation regarding mobile phones.

With a wider set of explanatory factors than prior studies, this one explores the factors that contribute to the formation of brand preference using a PLS model to understand the relationship between those and consumer preference for mobile phone brands.

Despite the exploratory nature of the study, the results reveal that brand identity, personality and image, together with self-image congruence have the highest impact on brand preference. Some other factors linked to the consumer and the situation also affect preference, but to a lesser degree.

The paper focuses on the application of a very common research issue in marketing: the analysis of the differences between groups’ structural relations. Although PLS path modeling has some advantages over covariance-based structural equation modeling (CBSEM) regarding this type of research issue – especially in the presence of formative indicators – few publications employ this method. This paper therefore presents an exemplary model that examines the effects of corporate-level marketing activities on corporate reputation as a mediating construct and, finally, on customer loyalty. PLS multi-group analysis is used to empirically test for differences between stakeholder groups in a sample from Germany’s mobile communications industry.

Reputation is one of the most important intangible assets of a firm. For the most part, recent articles have investigated its impact on firm profitability whereas its effects on individual customers have been neglected. Using data from consumers of an international consumer goods producer, this paper (1) focuses on measuring and discussing the relationships between corporate reputation, consumer satisfaction, and consumer loyalty and (2) examines possible moderating and mediating effects among the constructs. We find that reputation is an antecedent of satisfaction and loyalty that has hitherto been neglected by management. Furthermore, we find that more than half of the effect of reputation onto loyalty is mediated by satisfaction. This means that reputation can only partially be considered a substitute for a consumer’s own experiences with a firm. In order to achieve consumer loyalty, organizations need to create both, a good reputation and high satisfaction.

Customer value has received much attention in the recent marketing literature, but relatively little research has specifically focused on inclusion of service components when defining and operationalizing customer value. The purpose of this study is to gain a deeper understanding of customer value by examining several service elements, namely service quality, service equity, and relational benefits, as well as perceived sacrifice, in customer assessments of value. A multiple industry, cross-cultural setting is used to substantiate our inclusion of service components and to examine whether customer value is best modeled using formative or reflective measures. Our results suggest conceptualizing customer value with service components can be supported empirically, the use of formative components of service value can be supported both theoretically and empirically and is superior to a reflective operationalization of the construct, and that our measure is a robust one that works well across multiple service contexts and cultures.

Structural equation modeling (SEM) can be employed to emulate more traditional analysis techniques, such as MANOVA, discriminant analysis, and canonical correlation analysis. Recently, it has been realized that this emulation is not restricted to covariance-based SEM, but can easily be extended to components-based SEM, or partials least squares (PLS) path analysis (Guinot et al. 2001; Tenenhaus et al. 2005; Wetzels et al. 2005). In this paper, we will apply PLS path analysis to a fixed-effects, between-subjects factorial design in an online complaint-handling context. The results of our empirical study reveal that satisfaction with online recovery is determined by the level of both procedural and distributive justice. Furthermore, customers’ satisfaction with the way their complaints are handled has a positive influence on the customers’ intentions to repurchase and to spread positive word of mouth. Taking into account the entire chain of effects, we find that the influence of justice perceptions on behavioral intentions is almost fully mediated by satisfaction. From a managerial perspective, the results of our study provide insight into how to design effective complaint-handling strategies in order to maintain a satisfied and loyal customer base.

In an empirical study the strategies are investigated that content providers follow in their compensation policy with respect to their customers. The choice of the policy can be explained by the resource based view and may serve as recommendations. We illustrate how a strategy study in marketing can be analyzed with the help of PLS thereby providing more detailed and actionable results. First, complex measures have to be operationalized by more specific indicators, marketing instruments in our case, which proved to be formative in most cases. Only by using PLS it was possible to extract the influence of every single formative indicator on the final constructs, i.e., the monetary form of the partnerships. Second, PLS allows for more degrees of freedom so that a complex model could be estimated with a number of cases that would not be sufficient for ML-LISREL. Third, PLS does not work with distributional assumptions while significance tests can still be carried out with the help of bootstrapping. We recommend the use of PLS for future strategy studies in marketing because it is possible to extract the drivers at the indicator level so that detailed recommendations can be given for managing marketing instruments.

Advances in structural equation modeling (SEM) techniques have made it possible for management researchers to simultaneously examine theory and measures. When using sophisticated SEM techniques such as covariance-based structural equation modeling (CBSEM) and partial least squares (PLS), researchers must be aware of their underlying assumptions and limitations. SEM models such as PLS can help total quality management (TQM) researchers achieve new insights. Researchers in the area of TQM need to apply this technique properly in order to better understand the complex relationships proposed in their models. This paper attempts to apply PLS in the area of TQM research. Consequently, special emphasis is placed on identifying the relationships between the most prominent TQM constructs and business performance based on a sample of SMEs operating in the Turkish textile industry. The analysis of PLS results indicate that a good deal of support is found for the proposed model where a satisfactory percentage of the variance in the dependent constructs is explained by the independent constructs.

This chapter illustrates how PLS can be used when investigating causal models with moderators at a higher level of abstraction. This is accomplished with the presentation of a marketing example. This example specifically investigates the influence of brand personality on brand relationship quality with involvement being a moderator. The literature is reviewed on how to analyze moderational hypotheses with PLS. Considerable work is devoted to the process undertaken to analyze higher order structures. The results indicate that involvement does moderate the main effects relationship between brand personality and brand relationship quality. This chapter makes a unique contribution and applied researchers will appreciate the descriptive way it is written with regards to analytical process.

The objective of this paper is to provide a basic framework for researchers interested in reporting the results of their PLS analyses. Since the dominant paradigm in reporting Structural Equation Modeling results is covariance based, this paper begins by providing a discussion of key differences and rationale that researchers can use to support their use of PLS. This is followed by two examples from the discipline of Information Systems. The first consists of constructs with reflective indicators (mode A). This is followed up with a model that includes a construct with formative indicators (mode B).

This paper gives a basic comprehension of the partial least squares approach. In this context, the aim of this paper is to develop a guide for the evaluation of structural equation models, using the current statistical methods methodological knowledge by specifically considering the Partial-Least-Squares (PLS) approach’s requirements. As an advantage, the PLS method demands significantly fewer requirements compared to that of covariance structure analyses, but nevertheless delivers consistent estimation results. This makes PLS a valuable tool for testing theories. Another asset of the PLS approach is its ability to deal with formative as well as reflective indicators, even within one structural equation model. This indicates that the PLS approach is appropriate for explorative analysis of structural equation models, too, thus offering a significant contribution to theory development. However, little knowledge is available regarding the evaluating of PLS structural equation models. To overcome this research gap a broad and detailed guideline for the assessment of reflective and formative measurement models as well as of the structural model had been developed. Moreover, to illustrate the guideline, a detailed application of the evaluation criteria had been conducted to an empirical model explaining repeat purchasing behaviour.

Along with the development of scientific disciplines, namely social sciences, hypothesized relationships become increasingly more complex. Besides the examination of direct effects, researchers are more and more interested in moderating effects. Moderating effects are evoked by variables whose variation influences the strength or the direction of a relationship between an exogenous and an endogenous variable. Investigators using partial least squares path modeling need appropriate means to test their models for such moderating effects. We illustrate the identification and quantification of moderating effects in complex causal structures by means of Partial Least Squares Path Modeling. We also show that group comparisons, i.e. comparisons of model estimates for different groups of observations, represent a special case of moderating effects by having the grouping variable as a categorical moderator variable. We provide profound answers to typical questions related to testing moderating effects within PLS path models:

1.

How can a moderating effect be drawn in a PLS path model, taking into account that the available software only permits direct effects?

2.

How does the type of measurement model of the independent and the moderator variables influence the detection of moderating effects?

3.

Before the model estimation, should the data be prepared in a particular manner? Should the indicators be centered (by having a mean of zero), standardized (by having a mean of zero and a standard deviation of one), or manipulated in any other way?

4.

How can the coefficients of moderating effects be estimated and interpreted?And, finally:

5.

How can the significance of moderating effects be determined?

Borrowing from the body of knowledge on modeling interaction effect within multiple regression, we develop a guideline on how to test moderating effects in PLS path models. In particular, we create a graphical representation of the necessary steps to take and decisions to make in the form of a flow chart. Starting with the analysis of the type of data available, via the measurement model specification, the flow chart leads the researcher through the decisions on how to prepare the data and how to model the moderating effect. The flow chart ends with the bootstrapping, as the preferred means to test significance, and the final interpretation of the model outcomes.

After years of stagnancy, PLS path modeling has recently attracted renewed interest from applied researchers in marketing. At the same time, the availability of software alternatives to Lohmöller’s LVPLS package has considerably increased (PLS-Graph, PLS-GUI, SPAD-PLS, SmartPLS). To help the user to make an informed decision, the existing programs are reviewed with regard to requirements, methodological options, and ease-of-use; their strengths and weaknesses are identified. Furthermore, estimation results for different simulated data sets, each focusing on a specific issue (sign changes and bootstrapping, missing data, and multi-collinearity), are compared.

SIMCA-P is a kind of user-friendly software developed by Umetrics, which is mainly used for the methods of principle component analysis (PCA) and partial least square (PLS) regression. This paper introduces the main glossaries, analysis cycle and basic operations in SIMCA-P via a practical example. In the application section, this paper adopts SIMCA-P to estimate the PLS model with qualitative variables in independent variables set and applies it in the stand storm prevention in Beijing. Furthermore, this paper demonstrates the advantage of lowering the wind erosion by Conservation Tillage method and shows that Conservation Tillage is worth promotion in Beijing sand storm prevention.

A change in the behavior of the automotive customers has been noticed throughout the last years. Customers feel a renewed interest in the intangible assets of perceived quality and comfort of environment. A concrete case of study has been set up to analyze the preferences for 15 air conditioning supports. Descriptive data obtained by flash profiling with five experts on the photographs of 15 air conditioning supports are treated by Generalized Procrustes Analysis (GPA). The preferences of 61 customers are then explained by Partial Least Squares (PLS) regression applied to the factors selected from the GPA. The results provided by the XLSTAT GPA and PLS regression functions help to quickly identify the items that have a positive or negative impact on the customers’ preferences, and to define products that fit the customers’ expectations.