A STUDY ON PARTIAL LEAST SQUARES STRUCTURAL EQUATION MODELING (PLS-SEM) AS EMERGING TOOL IN ACTION RESEARCH

— Structural equation modeling (SEM) depicts one of the most salient research methods across a variety of disciplines, including educational management. Recent research advocates the use of partial least squares structural equation modeling (PLS-SEM) as an attractive tool in action research. The purpose of this paper is to systematically examine how PLS-SEM has been applied in action research with the aim of investigating the effect of teacher’s leadership styles (transformational and transaction) and student’s motivational factors (intrinsic and extrinsic) on student engagement for implementation of Communicative Language Teaching (CLT) in classrooms as well as explore the mediating role of motivational factors of student between the relationship of teacher’s leadership styles and student engagement. A cross-sectional survey design was used for the study featuring a self-administrated questionnaire among the students of some selected schools in Bangladesh. The transactional leadership style of teachers influences student engagement, intrinsic, and extrinsic motivation while the transformational leadership style of teachers influences the intrinsic and extrinsic motivation of the student. Both motivational factors mediated the relationship between leadership styles and student engagement. This study contributes to the literature by providing teachers with the updated guidelines for action research by using PLS-SEM. The study also suggests the way for increasing student engagement for CLT implementation in classrooms.


I. INTRODUCTION
To test whole theories as well as concepts, the acceptance and recognition of structural equation modeling (SEM) has developed (Rigdon, 1998). Copious of SEM's accomplishment can be accredited to the technique's capability to assess the measurement of latent variables whereas correspondingly testing connections amongst latent variables (Babin et al., 2008). Formerly established through Wold (1974Wold ( , 1980Wold ( , 1982 PLS is an SEM procedure grounded on an iterative method that take advantage of the elucidated discrepancy of endogenous constructs (Fornell and Bookstein, 1982). Essentially, PLS-SEM functions considerably similar to a multiple regression analysis (Hult et al., 2018). In around 1940s and 50s action research introduced by famous Kurt Lewin and his associates as a cooperative problem resolving sequence for refining administrations ( Lewin 1947( Lewin , 1948Corey 1953). The expression action research, apprehended the conception of well-organized examination in the circumstance of concentrated determinations to progress the excellence of an institute in addition to its action. Nowadays, action research vestiges as an authoritative instrument for instantaneously refining the practice and the well define structure of an institution.

II. LITERATURE REVIEW
management (Kaufmann and Gaeckler, 2015); marketing (Hair et al. 2012b); strategic management (Hair et al. 2012a); management information systems (Ringle et al. 2012); operations management (Peng and Lai, 2012) and accounting (Lee et al. 2011). Over-all 875 research have been directed grounded on PLS-SEM from the period of 1980 to 2017 conferring to . No research has been directed based on action research with PLS-SEM. The all-inclusive learning groups can be invigorate by the using of Action research, in addition to it assistance educators in altering or reproducing on their classroom practices. This can assist resourcefulness of every individual educators, educational institutions, and institutes functioning with societies, as well as regions, districts. The PLS -SEM method has incredible capability to manage the problematic modeling issues that routinely take place in the social sciences such as unusual data characteristics like non-normal data and highly complex models. Consequently, published research has frequently accentuated that PLS -SEM is predominantly engaging for applied science by consenting the testing of hypothesized connections in taking a extrapolation or prediction emphasis in the model estimation (Evermann and Tate, 2016;Sarstedt et al., 2017). PLS -SEM accordingly incapacitates the obvious dichotomy amongst clarification, which academic research generally accentuates, in addition to prediction which is obligatory to derive managerial implications (Hair et al., 2019). Furthermore, different controversial views about the merits and demerits of method, witnessed in various fields of research (Khan et al., 2019), have enhanced the understandings of it (Petter, 2018).

Action Research for Communicative Language Teaching
Implementation CLT syllabus emphasis on communication which encourages students to learn a language better with the meaningful prospect and change the traditional classroom practice. Language management theory argues to address that language planning at first must find out the problems in the related context, and the planning process must solve all the problems as well as give suggestions to manage every aspect to complete the planning (Neustupný, 1994). According to the theory if hindrances persist then language implementation is not possible. So CLT implantation of Bangladeshi Secondary Schools is not possible if the problems like minimum motivation of the students and inappropriate leadership style of the teachers exist.
Using PLS-SEM, this study is going to investigate: (1) The effect of Teacher's leadership style (transformational and transactional) on student's motivation (intrinsic and extrinsic) in classroom.
(2) The influence of Teacher's leadership (transformational and transactional) on student's engagement in classroom.
(3) Student motivation (intrinsic and extrinsic) plays mediating role between the relationships of teacher's leadership style (transformational and transactional) and student's engagement in classroom.

WHY TO USE PLS-SEM FOR ACTION RESEARCH
The PLS-SEM methodology is used because of gaining the acceptance of many business discipline . Many scholars have published their papers by summarizing and using the PLS-SEM in the field of their research. This paper presented the three major reasons for applying PLS-SEM in the field of action research that are data distribution, sample size, and the usages of formative indicators (Lin, et al. 2020).

Non-normal data
Collection of data for action research often is unsuccessful to pursue multivariate normal distribution. PLS-SEM is flexible to work with non-normal data due to PLS algorithm's transformation of non-normal data in accordance with central limit theorem (Beebe et al., 1998;Cassel et al., 1999). Hence, the caveat to PLS-SEM is to provide the complete solutions to models by using the non-normal data is twofold.
First, the researchers need to be careful that the highly skewed data can weaken the statistical power of the analysis. More specifically it is said that the valuation of the model parameters' significance depends on normal errors from bootstrapping that may be inflated in case highly skewed data (Henseler et al., 2016). Second, because CB-SEM is concerned with many alternatives to estimate procedures that may be problematic causes of assuming the PLS-SEM when data distribution is the automatic choice (Hair et al., 2012b).

Small sample size
PLS-SEM is useful when it works with small sample sizes (Chin & Newsted, 1999). Though the sample size brings impact on the different aspects of SEM that contains parameter estimates, model fit, and statistical power (Shah and Goldstein, 2006).  presented that the PLS is distribution free and good for studying the difficult models that have sample sizes. PLS-SEM is able to achieve the higher rate of statistical power and shows the superior convergence behaviour (Henseler, 2010). A popular heuristic indicates that small sample size for PLS model should be equal to ten times biggest number of formative indicators that are used to detect one construct or ten times the biggest number of inner model paths conducted at a particular construct in the inner model (Barclay et al., 1995).

Formative indicators
There is a difference between reflective and formative constructs. The formative measures shows the cases where the indicators cause the construct (i.e. the arrows point from the indicators to the construct), On the other hand construct causes the reflective indicators (i.e. the arrows point from the construct to the indicators).Hence, the PLS-SEM and CB-SEM are able to measure the models by using formative indicators, PLS-SEM has gained the mentionable support as the recognized method . Formative indicators with CB-SEM creates the problems of identification while being analyzed (Jarvis et al., 2003), it is common to researchers in believing that PLS-SEM is the finest option.

III. METHODOLOGY
A cross-sectional survey design was used for the study featuring a self-administrated questionnaire. The quantitative research design is applied and the survey instrument for the study was questionnaire. The population was the students of some selected higher secondary schools in Bangladesh. Partial random sampling had followed because it is the modified version of simple random sampling where researchers focused on every subgroup (every class) of the given population. The total sample was 387. In this present study, every construct in the questionnaire has three or more items where responses would be elicited using the Five-Point Likert Scale.

IV. DATA ANALYSIS & RESULTS
The researchers are required to follow the multi stage-process that is concerned with specification of the inner and outer models, data collection and assessment, the actual model estimation, and the valuation of results while applying the

PLS-SEM.
This study is centered by the three major steps that are given below: (1) Model specification; (2) Outer model evaluation; and (3) Inner model evaluation.

Hair et al. (2014) presents the details introduction into every
stages of PLS-SEM use.
(1) Model specification The model specification stage involves with setting up of the inner and outer models. The inner model or structural model shows the interactions between the constructs being appraised.
The outer models are also acknowledged as the measurement models that are used in evaluating the interactions between the indicator variables and their corresponding construct.
On the basis of theory and logic the first step of using PLS-SEM is concerned with creating a path model which connects variables and constructs . In preparing the path model which is shown in Figure  (2) Outer model evaluation 134 between reflectively and formatively measured constructs after apprising the outer models Sarstedt and Schloderer, 2010). Different concepts are based on the two approaches to measurement. Hence, it needs the consideration of various evaluative measures.

Reflective indicators
PLS-SEM include two types of measurement model one is reflective and another is formative measurement model.
Therefore, the researchers differentiate between these two types of models to assess them (Henseler, Ringle, and Sinkovics 2009 (Henseler, Ringle & Sarstedt, 2015). Furthermore, PLS-SEM is a non parametric bootstrap method that makes no distributional assumption and can be estimated with small sample sizes (Hair, Hult, et al., 2017). Hence, PLS-SEM is not concerned with standard goodness-of-fit statistic and prior efforts of establishing a matching statistic that has proved highly problematic (Henseler and Sarstedt, 2013

Coefficient of determination (R square)
The R square quantifies the predictive accuracy of model. On the other hand R square explains the exogenous variable's united effect on the endogenous variable(s). This effect presents the ranges from 0 to 1 where 1 shows the total predictive accuracy. As R square embraces different disciplines, scholars relies on a "rough" rule of thumb considering an standard R square, with 0.75, 0.50, 0.25, respectively. It indicates the large, modest, or weak levels of predictive accuracy (Hair et al., 2011;Henseler et al., 2009

Cross-validated redundancy (Q square)
The Q square is a means that measure the predictive relevance of inner model. The assessment builds on a sample re-use technique where it skip over a part of the data matrix, assess the parameters of model and predicts the skipped element with uses of estimates. The smaller difference between predicted and original values indicates the larger Q square and thus predictive accuracy of the model. More distinctively, the value of Q square is larger than zero of particular endogenous construct which shows the predictive relevance of path model for this particular construct (Table 7). Therefore it should be mentioned that by comparing the value of Q square to zero is investigative whether an endogenous construct can be predicted, it does not mean anything about the quality of the prediction (Rigdon, 2014;Sarstedt et al., 2014).  (Cohen, 1988). to -1or + 1 are statistically significant always, a standard error is attained with the uses of bootstrapping to examine the significance (Helm et al., 2009). After measuring the significant relationships the researchers need to justify the relevance of significant relationships. In a brief are the sizes of the structural coefficients meaningful? As stated by Hair et al. (2014), many studies overlook this step and merely rely on the significance of effects. If this important step is omitted, researchers may focus on a relationship that, although significant, may be too small to merit managerial attention. If the p value is less than 0.05 and the t-value is higher than 1.96 then the effect will be significant at confidence level 95%.  2; While, transformational leadership style has no effect on student engagement (b=0.034, t= 0.532, p= 0.601). and student engagement is sequentially mediated by student motivation (intrinsic and extrinsic). As such, this analysisopposed to a simple evaluation of direct effectsprovides a more appropriate picture of action research performance.
Several authors have criticized the far-reaching neglect of explicitly examining mediating effects in PLS path models, which can easily lead to erroneous conclusions when interpreting model estimates (Hair et al., 2013(Hair et al., , 2012a. A potential reason for this neglect might be that there is still some ambiguity on how to evaluate mediating effects in PLS-SEM. Hair et al. (2014) provide an initial illustration on how to analyze mediating effects. According to the

VII. CONCLUSION AND LIMITATION
In conclusion, in addition to effects on instructional activities, action research enables teachers to work as researchers, too.
As practitioners of curriculum and educational programs teachers are also a variable in education settings (like, CLT implementation). The teaching process is a dynamic, humane