Quantitative research approach Professor Yvonne Brunetto Professor Stephen

Quantitative research approach Professor Yvonne Brunetto, Professor Stephen Teo, Professor Jarrod Haar

Three basic criteria determines which method to use 1. The type of research question. E. g “Why” and “How” questions are ideally suited to a case study approach. 2. The extent to which the researcher has control over the subject and context of the subject e. g. , The experimental approach is more suitable for controlled environments, whereas the case study approach is better if you can’t control subjects or the context 3. The period of history under examination influences the choice of methodology. E. G if you want to study an historical phenomena , then you have to use archives information

The Need for a Framework The conceptual/theoretical framework is a logically developed, described and elaborated network of associations among concepts or variables deemed relevant to the problem situation.

The basis of quantitative research: Central Tendency Across any one SAMPLE there will typically be a standard distribution of a particular property, for example it is said that intelligence across a country’s population will adhere to a standard distribution. F r e q u e n c y Populations and samples that adhere to a standard distribution will have a majority of participations existing in the centre of any distribution – and this frequency will decrease as the deviations get closer to the extremes at any end of the distribution

Theoretical Framework 1. Types of variables • • Dependent Independent Moderating Intervening 2. Components of Theoretical Framework

Concepts and Variables A Concept • an idea expressed as a symbol or in words • Conceptual framework A Variable • Can be observed and measured • Theoretical framework

Dependent variable • A dependent variable is a measurable outcome of an experiment. – For example communication satisfaction, productivity, number of sick days, employee morale, or organisational commitment could be dependent variables. There is a clear advantage if the dependent variable is easily measured; sick days are much easier to measure than employee morale.

Independent variable(s) • An independent variable or treatment variable represents a quality or characteristic that is varied or manipulated during the experiment. – examples include quality of feedback, training methods, remuneration, or work hours. The independent or treatment variable is manipulated to determine the effect on the dependent variable. This process is the treatment received by the participants. The group receiving the treatment is referred to as the treatment or experimental group.

Dr Apivut Chakuthip thesis

Theoretical Framework Independent variables • POS (Perceived organisational support) Social Exchange Theory • Procedural justice • Interactional justice • LMX • Trust • Organisational culture • Tie Strength Dependent Variables Intention to turnover of nursing professionals Affective Commitment OCB Innovative Behaviour Dr Matt Xerri ‘s Theoretical framework

Independent Variables Training and Development Communication Processes: • Frequency • Informal • Indirect • Two-way Moderator Ambiguity Regarding Customers Dependent Variables Job Satisfaction Employee Performance An example of a conceptual framework Affective Commitment Professionalism Dimensions: • Referent • Self-Regulation • Autonomy Organisation Occupation Customers By Dr Natasha Currant

Using SEM for your research Professor Stephen Teo Management Department AUT Business School

What is SEM [structural equations modelling]? q A technique for testing theoretical models q Researcher specifies their model and how the various constructs should influence each other q Hoyle’s (1994) review tells us that SEM can address: q Questions about causal process q Basic questions of measurement q Questions about causal process when variables are not well measured q SEM methods share most of the strengths of OLS multiple regression q SEM tests this model statistically q Incorporates the features of factor analysis and regression analysis

Strengths q Proposed causal explanations are made explicit q Tests of fit allow implausible models to be rejected q Competing models can often be compared, and one may emerge as more plausible given the data. Limitations q Models are often mis-specified (needs theorizing): q Linearity assumption is often made uncritically q Measurement error distorts analysis q Important variables may be missing q Communicating results is challenging q Novices may overstate claims or make errors in complex analyses that are difficult to detect

Covariance-based SEM q Question to ask during the modelling process q Could this model have led to the data that I have? q That’s why in AMOS, Mplus and even PLS, we refer to goodness of fit indices to show model fit (eg. χ2/df=2. 076, CFI=. 95, TLI=. 94, RMSEA=. 04, SRMR=. 07 Model Data

Figure 2 a. Results of Analysis using Mplus (Female) WLB . 09** Info Provision (. 12*) AC R 2=41. 3% (. 06*). 23*** Involvement . 55***. 31*** Job Satisfaction . 29*** Influence Goodness of fit: 2 /df=1. 77, RMSEA=0. 032, CFI=0. 991, TLI=. 988, SRMR=0. 021 Source: Ravenswood and Teo (2014) in AIRAANZ Conference

• As Yvonne mentioned, “Theory” is important and SEM, is a theory driven process – Theory is specified as a model • Alternative theories can be tested – Specified as models (Q: simple or complex? ) Theory B Theory A Data

Ambidex Strg Mgt Org Sys Org Perf HRM System Work Attitudes Strg HRM Source: Plimmer, Teo and Bryson (2014): using PLS

Partial Least Squares (PLS) Modelling q PLS is a latent path model, a well-established technique for estimating path coefficients in causal modelling q Statistical basis initially formed in the late 60 s through the 70 s by econometricians in Europe q The conceptual core of PLS is an iterative combination of principal components analysis relating measures (each questionnaire item) to constructs (latent variables or factors), and path analysis permitting the construction of a system of constructs q Allows for the simultaneous testing of hypotheses, unlike multiple regression, within the same statistical analysis

Structural (path) model (lines and circles) 2 nd order LV 1 st order LV Measurement (items within each construct) model (blue)

Positives of PLS Modelling q Does not require normal data q Accept smaller sample sizes because “each causal subsystem sequence of paths is estimated separately. … and is particularly suitable for studies in the early stages of theory development and testing…” (Johansson & Yip, 1994, 587) q. Min sample size 30 to 100 (Chin and Newstead 1999), but Green (1991) has a set of ‘rule of thumb’ based on the number of ‘independent variables’ (predictors) in the model q Combined regression and factor analysis within the model (measurement model) in each “run” q Suitable for developing constructs and models for further testing

Negatives of PLS Modelling q Does not allow conventional test for goodness of fit as per AMOS q Unable to test for co-variance relationships between variables (constructs) q Not suitable for testing theory (and model) q Irrespective of which technique (AMOS or PLS), researchers must consider the threat of Common method bias (Podsakoff et al. , 2003) post-hoc (Harman’s one factor test) or using method factor (see Rafferty & Griffin, 2004) or using longitudinal data (see example in Teo et al. , 2013)

Time 1 Time 2 Source: Teo et al. (2013) Journal of Nursing Management

Multi-Level Analysis Professor Jarrod Haar School of Management Massey University (Albany)

What is Multi-Level? • Multi-level represents a different level of analysis • Single source data is typically analysed with more ‘simply analysis’ ie regression in SPSS, to SEM in AMOS or Mplus • Fundamentally, the relations are on the same level ie an employee with more work-life balance has more job satisfaction. Thus,

Simple Linear Regression

Why Multi-Level? • For when we explore relationships that are NOT on the same level • This is because data is ‘nested’ e. g. , it might be nested in teams, or follower data is nested under a specific leader • This ‘nesting’ effect means ‘Simple Linear Regression’ is not sufficient – as it does not pick up and account for these ‘nested’ effects… thus, multi-level analysis might look like…

Multi-Level Linear Regression

Multi-Level? • With the advances in technology (statistical programs) we can now readily conduct multi-level studies where previously these have been particularly difficult. • Programs such as Mlwi. N, Mplus, and HLM • These approaches are advantageous (from a theoretical contribution, empirical contribution) and thus are more readily received in journal publishing. But, the data is typically harder to get!

Multi-Level Examples • So, what would these relationships look like? • Well, in the Pure Sciences it might go from: cell to organ to person to a family [thus: cell=neurochemistry; organ=ability to metabolize ethanol; person=genetic susceptibility to addiction; family=alcohol abuse in the home] • In the Social Sciences (Management) we might explore: employees – teams – departments – divisions – organisational sites – firms in industries – etc…

Multi-Level Examples • Specific management examples: • Teams (collective) IVs (e. g. , safety climate) to individual outcomes (e. g. , OCBs, Job Sat, Turnover Intentions) • Individuals (e. g. , personality styles) to Team Outcomes (e. g. , team performance, team wellbeing) • Leaders influence (e. g. , transformational leadership style) to Team Outcomes (individual or collective/team)

Multi-Level Summary • Fundamentally, the issues and rules behind good quantitative research simply apply to multi-level research. But good research needs context specific analysis – team data in SEM is not right! • The data collection might be more onerous (e. g. , team data is hard – mean majority typically!); and time demands (e. g. , takes longer); issues of trust (e. g. , leaders and their followers)… • The trade off? You can get away with smaller sample sizes e. g. Spell et al. (2011) SGR (ABDC=A) had 42 teams (n=174 employees)