According to Hair et al. (2010:7), the researcher’s objective of reducing measurement error can follow several paths, and the researcher must address two important characteristics of a measure, which includes the face validity of the questionnaire, and the reliability of the questionnaire to produce consistent results.
4.8.1 Validity
Validity is related to the test measures that the researcher actually wishes to measure; it is the level to which the researcher measures the accuracy of representations compared to what it is supposed to be. Ensuring validity starts with a thorough understanding of what is to be measured and then, ensuring that the measurement is correct. However, accuracy does not assure validity (Hair et al., 2010:7).
The face validity of the questionnaire is tested using a pilot study to assess the quality of the questionnaire regarding its ability to achieve the research goals. Pre-testing of questions was done to make certain that the respondents understood the questionnaire and had furnished the right responses. Pretesting also ensured relevance, effectiveness and clarity of the questionnaire. An initial group of Ten SMEs was selected using a randomly selected sample from the population of the Pietermaritzburg Chamber of Business. Pilot study was conducted in the first week of August 2016 and the data from the pilot study were not included in the study.
The statistician advised on whether the expected output from each question was within the research parameters, and the expert’s feedback was used to modify the questionnaire where appropriate and necessary. The questionnaire was designed shortly and succinctly to prevent misinterpretation and any ambiguity. The language was simple and easy to understand. To ensure validity, questions were adjusted by responses and comments received from the pre-test.
The study assessed both convergent and discriminant validity. Convergent validity results will be provided first and then followed by discriminant validity. According to Hair, Black,
Babin, Anderson and Tatham (2006:771) convergent validity is “the extent at which indicators of a specific variable converge or share a high proportion of variance in common.” It simply explains the extent at which a scale correlates with other measures of the same construct to the same direction. According to Carlson and Herdman (2012), weaker convergent validity is evident using values deviating from one while values closer to one are normally accepted.
The average variance extracted approximately reflects the total elements of variance in the indicators which are accounted for by a latent construct. Dillon and Goldstein (1984) suggested that an AVE value greater than 0.50 demonstrates that the convergent validity of the variable is good. According to Fraering and Minor (2006), an AVE value of 0.4 is seen as satisfactory. The values were calculated using Amos software and were cross- checked again with the manual calculation, which resulted in same values using the formula below:
Vη=Σλyi2/ (Σλyi2+Σεi)
AVE = {(summation of the squared of factor loadings)/ {(summation of the squared of factor loadings) + (summation of error variances)}.
Kline (2011) defined discriminant validity as contrary to convergent validity in the extent that variables alleged to evaluate different variables shows discriminant validity. Discriminant validity describes how measures in the same study are distinct from other measures. Inter-construct correlation matrices and Average variance extracted (AVE) compared to Shared variance (SV) were used to assess the discriminant validity in the current study.
The discriminant validity of the study was examined through an examination of the correlation values of the research constructs. The correlation values range from 0 – 1. A low correlation between research constructs indicates that the research constructs are unique and distinct from one another – while the reverse indicates the absence of discriminant validity. Theoretically, a correlation value less than 0.6 is deemed an indicator of discriminant validity. However, practically, a correlation value that is less than 0.85 is still regarded marginally acceptable (Chinomona, 2011).
4.8.2 Reliability
Reliability refers to the accuracy and precision of a measurement process (Hair et al., 2010), and if the validity is assured, the researcher must still consider the reliability of the measurements. According to Hair et al. (2010:7), reliability is the level to which the researcher measures the true value of the observed variable, and that it is error free. The reliability of the measurement instrument was assessed using both the Cronbach alpha coefficients and Composite Reliability indicators. Cronbach’s coefficient alpha was used to evaluate the measurement scale in the study. Thus, Cronbach’s alpha was used to verify the internal consistency of the variables – that is, to evaluate the reliability of the measurements of each variable. According to Kipkebut (2010) values for Cronbach alpha ranges between 0 and 1. Hair et al. (2009) also indicated that “values higher than 0.6 were considered as being reliable”.
The Composite Reliability (CR) index was also used to check internal consistency of the
measurement model. Ramayah et al. (2011) observed that, composite reliability shows
the extent to which variable indicators identify the latent variable. Urbach and Ahlemann (2010) posited that values that are acceptable are normally between zero and one. According to Vicente, Abrantes and Teixeira (2015), it is recommended that composite reliability values must exceed 0.7. The composite reliability test was calculated using the following formula:
CRη = (Σλyi)2/ [(Σλyi)2+ (Σεi)]
Composite Reliability = (square of the summation of the factor loadings)/ {(square of the summation of the factor loadings) + (summation of error variances)}.
4.8.3 Ethical Considerations
Garrard and Narayan (2013) suggested that ethical considerations are critical issues to be considered in research. In fact, for human participation in a research study, it is ethically required that the researcher should obtain an informed consent from the participants. Thus, the consent form (see appendix 10) was sent to the participants as an attachment to the questionnaire, and the participants who were willing to participate in the survey had to read, sign and return the consent forms together with the completed
questionnaires. Ethical clearance was obtained from the University of KwaZulu-Natal’s Research Ethics Committee in May 2016, to conduct the study.