Data Analysis with SPSS for Survey-based Research

By Saiyidi Mat Roni and Hadrian Geri Djajadikerta

This book is written for research students and early-career researchers to quickly and easily learn how to analyse data using SPSS. It follows commonly used logical steps in data analysis design for research. The book features SPSS screenshots to assist rapid acquisition of the techniques required to process their research data.

Rather than using a conventional writing style to discuss fundamentals of statistics, this book focuses directly on the technical aspects of using SPSS to analyse data. This approach allows researchers and research students to spend more time on interpretations and discussions of SPSS outputs, rather than on the mundane task of actually processing their data.

• Language: English
• Publisher: Springer
• Release date: Jun 21, 2021
• ISBN: 9789811601934

Book preview

© Springer Nature Singapore Pte Ltd. 2021

S. Mat Roni, H. G. DjajadikertaData Analysis with SPSS for Survey-based Researchhttps://doi.org/10.1007/978-981-16-0193-4_1

1. Research Instrument Design and Sample

Saiyidi Mat Roni¹   and Hadrian Geri Djajadikerta¹

(1)

School of Business and Law, Edith Cowan University, Joondalup, WA, Australia

Abstract

Research is a journey. While the destination is the end, it is the process that adds colours to the fun. A good research answers questions and offers solutions. Even when the research concludes with no definitive answer, the work offers ways to investigate a phenomenon (or demonstrates the not-to-do ways because these are deemed unsuccessful!). And to kick off with the journey, we explore two technical parts of research in this chapter– survey questionnaire and sample size. The survey questionnaire is the tool where you use to collect data from the sample. The sample size is the total number of individuals for whom the survey is sent. We demonstrate how you can create the right questions to get the right answers from the right number of individuals.

Keywords

Latent variableQuestionnaireSample sizeStatistical powerSurvey

Throughout our experience in research and supervising research students, two questions we are always consulted on after getting the research framework done are (1) how to collect the data, and (2) how many individuals do we need. Short answers for these questions are (1) design a survey, and (2) more is better. These do not answer the questions but they provide some hints on what to do next and a glimpse into how much more work is needed. Don’t panic. We will walk you through the process in this chapter. Specifically, we will discuss the design of a survey in relation to the variables, and the sample size to provide more contexts of answer 1 and 2.

1.1 Survey

A survey based research, as the name speaks, involves a survey instrument. Designing a survey takes time. There are other aspects of a research that intertwine into a survey design apart from the questions we need to meet research objectives. Smith (2015) in his book, Research Methods in Accounting, outlines seven aspects of a survey that needs to be addressed at the planning and design stage. These are the type of survey either mail, email, phone, or in-person; the sort of respondents such as professionals, or students; the response categories such as opinions, judgments, or knowledge; the sequence of the questions because it may influence the responses, hence biasing the results; the length of the survey as too short may not be sufficient to meet the objectives of a research, or too long that we can end up with many unanswered questions; the sample size (as expected); and of course, the questions which have to be carefully worded.

A good survey instrument must sufficiently and accurately measure the variables of interest, and also tap into the dimensions of the variables you are investigating. The former defines what is needed when you are collecting data for directly measured variables, and the latter is more or less like looking at beacons that make up an abstract, conceptual variable .

To give a better idea what we mean above, let’s assume you want to investigate a relationship between work experience (let’s call this experience) and job satisfaction (we call this satisfaction) of employees working at an organisation. Your survey instrument should include a question on the number of years working at the organisation to measure experience, and a series of questions on job satisfaction such as how satisfied the employees are, how happy they are, and how good they feel about working at the firm. Experience is collected through one single question because this is a directly measured variable . Satisfaction on the other hand, is a bit complex to gauge. It is an abstract and subjective variable . Therefore, we need a set of questions each tapping into various facets of satisfaction – the overall satisfaction (i.e., how satisfied they are), happiness, and feeling good. These are three dimensions of job satisfaction. A note on this though – this is only an example. You will need a good theoretical framework that defines job satisfaction in order to appropriately determine what constitute as dimensions for job satisfaction.

1.1.1 Working with Directly Observed Variables

A directly observed variable is, well, a variable that you can easily quantify. Demographic variables such as age, occupation, gender, income, and experience such as in the prologue above, can be directly quantified. One question is normally sufficient. Obviously, you don’t ask a person three different ways to know his or her age! But one thing you need to think of is that the way the response is extracted from participants of your study can have a profound effect on the way and the type of statistical analysis you will use. Take the following three survey instrument samples where each asks the same thing – the age of the respondent.

../images/466907_1_En_1_Chapter/466907_1_En_1_Figa_HTML.png../images/466907_1_En_1_Chapter/466907_1_En_1_Figb_HTML.png../images/466907_1_En_1_Chapter/466907_1_En_1_Figc_HTML.png

As you can see, although measuring a directly observable data such as age is a straightforward, single question, the way we ask and the response options we present can impact the participant’s willingness to respond, change the data type, and limit or expand the options for statistical analysis. Therefore, you should always take into account the tests that you need to run to meet your research objectives when designing your survey instrument.

1.1.2 Designing Questions for Latent Variables

By this stage you should be able to distinguish between directly observable and latent variables. The directly observable variable typically is the one that can be measured in most part through a single question. Age and income in the previous section are two examples of this variable type.

A latent variable on the other hand, is a variable that is observed (or measured) through or by its indicators. For example, attitude and trust are not directly observable, but can be estimated through a series of questions which taps into different facets of these variables. The challenge is to find out what the facets are (we use the word challenge instead of problem. It sounds more positive and encouraging). Some researchers take a smart shortcut – dig up the literature for well-known, highly cited question-items and use these items in their survey instrument. For instance, we can use the well-developed questions by Ajzen (1991) to measure attitude. The author is an expert in the Theory of Planned Behaviour (TPB) where attitude is one of the predictors in the TPB model. This saves time (by a lot)!

However, in some cases, a ready-made instrument is hard to come by. We need to develop the questions on our own. For this, we need to define the latent variable and draw in information from existing studies. Let’s get back to measuring variable trust. In order to tap into the correct dimensions, we first examine trust definitions from the following studies.

1.

Trust is one’s willingness to rely on another party in whom one has confidence (Moorman et al. 1993).

2.

Trust is one’s confidence in an exchange party’s reliability and integrity (Morgan and Hunt 1994).

3.

Trust is one’s willingness to be vulnerable to the action of the other party (Rousseau et al. 1998).

The core elements of trust based on these existing works are (1) confidence, (2) reliability, (3) integrity, and (4) willingness to be vulnerable (let’s make this simple by calling it willingness to accept risk). We can then deduce our definition of trust which encompasses these four facets of trust. Our definition can be written like this.

Trust is the level of confidence one has in other party’s reliability and integrity; and therefore, willingly accepts the risks associated with one’s action.

We have to say that this definition is not our best. But it serves the purpose, nevertheless. From this definition, we can then write a set of questions that taps into the dimension of trust. Here is an example of how the questions look like in the survey. You can see that each question asks a specific facet or dimension of trust. By the way, we add one more element, trustworthiness (see Lee and Turban 2001), as question 5 to measure overall trust.

../images/466907_1_En_1_Chapter/466907_1_En_1_Figd_HTML.png

1.2 Sample Size

Sample size is another interesting question and the one that we are always being asked. How many observations? It is easy to pick up a number, but the hard part is to provide a reasonable explanation for the sample size we choose. We use ‘reasonable’ rather than ‘correct’ as there are multiple justifications to decide how large the sample size should be. We encountered many instances where research students or researchers provided reasons for their decision on a sample size, but other ‘experts’ disagree in many ways. Sometimes the experts’ arguments against the sample size are detached from the reality and the context of the study. In one case, a research student was asked why he chose 120 respondents where an estimate from statistical power software showed it was 118. Well, it is obvious that in the sample size calculations of many software is based on the minimum sample size required for a stable statistical result, and applying the ‘more is better’ rule, choosing 120 is better. Anyway, 120 was above the minimum requirement.

Given the reliability and validity of a study is (partly) contingent to the sample size, the determination of it is therefore critical. Before anything else, let’s just come back to the purpose of sampling – cost and practicality. We have finite resources in term of time and money, among other things, when conducting research. Therefore, we take samples. And in order for the results of our analysis to be valid, reliable, and generalisable, we need to ensure that the samples are statistically good and representative of the population.

One technique that is commonly used to ensure the representativeness of the sample is to use random sampling. This is rather a straightforward method where each person in the population of interest has an equal chance to be picked up in the sample selection. But of course, if you are conducting a research to represent a population of a country, we encourage you to stratify the sample into sub-samples and randomly select the samples from each sub. For example, you can stratify your country-sample into regions and randomly distribute your survey to each region. This is called stratified random sampling technique. At the aggregate level, when you run an analysis, your result is more generalisable. Of course, you can also check if your result differs across multiple regions. If it does, then you can investigate regional variables that explain why people in different regions behave differently even though they are from the same country. This adds more story to your analysis.

Once you have determined the population and how to draw the sample, we need to determine the sample size. This is the answer to the ‘more is better’ rule. You can calculate the minimum sample size required for your analysis through a statistical power test. In this example, we demonstrate the use of G*Power (Faul et al. 2009) application that you can download from:

http://​www.​psychologie.​hhu.​de/​arbeitsgruppen/​allgemeine-psychologie-und-arbeitspsycholog​ie/​gpower.​html

At the time of writing this book, the application is available for free, and the website also provides a manual on how to operate G*Power. As a gesture of appreciation for their work, we highly encourage you to cite their work appropriately. You can find a full citation in the reference section of this book and of course, from their website.

Now, let’s have a look at an example on how to use G*Power to determine a minimum sample size for a given statistical test. In this example, we demonstrate the calculation for the minimum sample size for a correlation test using the application’s default parameter settings.

Launch the program and you will be presented with the main window.

Choose the main tab (Central and noncentral distributions).

Test family > choose Exact.

Statistical test > Choose Correlation: Bivariate normal model.

Type of power analysis > Select A priori: Compute required sample size – given α, power, and effect size. At this stage, you can leave the Input Parameters as default.

Click Calculate and you’re the minimum