Theoretical background of this study grounds on multiple research papers. The key concepts of the study were well-established in the research literature. The vast literature provides evidence for the hypotheses, and ought to back the findings of this Master’s thesis up. Strong references strenghten the reliability and validity of the study.
Reliability means, that the study is done so systematically, that anyone could repeat the study and get the same results. Validity means, that the study has answered the
research questions it originally was supposed to answer. (Metsämuuronen 2000, 11.) A good validity of a study means that the results may easily be generalizable (Metsämuuronen 2009, 65). A careful consideration of all of the steps of the study strenghten the reliability and validity of the study and the methods.
The question of, how IC assets and KM practices interact with each other and impact organizational performance ought to be answered with the chosen study method.
The mediation model has been very much used in psychology, but such a model is rare in a business management context. Discussion and previous research on mediation was found in the literature, and it supported the use of this model in studying the impacts of IC assets and KM practices on organizational performance.
Because the survey was peer-reviewed, or evaluated by IC and KM experts, and also pre-tested, the content validity can be said to be of high quality. However, the data was gathered from a single respondent per an organization. This means that the answers given were subjective in nature, and might be altered because of expectations of others. Therefore, there was a risk, that the answers were not reliable in all respects. For example simple misunderstandings or the respondents remembering something wrong, could weaken the reliability of a study. There is also a benefit of a doubt that the answers could be given wrong in purpose. To prevent intentional altering of the answers, the respondents were assured of confidentiality (Inkinen 2016). Respondents answering each question based on their own perception is a known limitation of practicly every quantitative research, and it needs to be accepted (Zack et al. 2009, 398). Because the survey used a five-point Likert scale, it also leaves a benefit of a doubt to a central tendency bias in responses.
This bias could have been avoided by using a six-point Likert scale (Andreeva &
Kianto 2012, 626).
The validity and reliability of the variables, measuring what they were supposed to measure, was confirmed quantitatively in factor analyses because the questions that intended to measure for example human capital created a factor of their own, the
questions intended to measure structural capital created a factor of their own, and so on. The pre-testing of the survey and the consulting of IC and KM experts also strenghten the validity of the study.
The reliability of the measures was quantitatively tested using Cronbach’s Alpha coefficient test. It tests, how well the measurement model works by splitting the observations into two groups that ought to measure the same thing even when separated. So to say, Cronbach’s Alpha tests the consistency of the data by analyzing correlation averages between the variables (F-Jardón & Martos 2009, 609; Metsämuuronen 2000b, 52). What needs to be known is that, standardized Alphas measure the dissimilarity of variances between variables differently than raw Alphas, and that they may be used in estimating changes in reliability in situations where the number of items in a scale or measurement varies. This was not the case in this study however.
Table 4. Cronbach’s Alpha coefficient test
Table 4. shows raw Alpha being 0.787, which marks the lowest limit of reliability of the model. The reliability is good, when Cronbach’s Alpha coefficient is higher than 0.7 (F-Jardón & Martos 2009, 609). A high Cronbach’s Alpha coefficient strenghtens the reliability of the study. A high reliability reveales, if the measures test similar features, and if respondents answered in a similar way to many questions.
(Metsämuuronen 2000b, 52.) Variables Alpha
Raw 0.787113
Standardized 0.813524 Cronbach Coefficient
Alpha
Table 5. Cronbach’s Alpha coefficient test when variables are deleted
Table 5. shows that 6 out of 7 variables resulted in raw Alphas between 0.72-0.77, if they were to be deleted from the study. Because Cronbach’s Alpha coefficient was 0.787, the only variable that would have resulted in a higher Alpha when eliminated, was compensation. Deleting this variable would have strengthened the reliability of the measures (Alpha 0.807). However, the Alpha was high to begin with and the variable was chosen to be included in the final measures. Again, the Alphas show the lowest limit of reliability of each variable in question. Table 5. also describes the total correlations that ought to be over 0.20 (Metsämuuronen 2009, 548). All of the variables showed values well above this limit (0.40-0.70).
The validity and reliability of the study and measures were considered and tested.
All of the analyses done before testing mediation strengthen the reliability and validity of the results, because the variables were critically observed, and the steps of the study carefully documented.
Correlation
with Total Alpha Correlation
with Total Alpha VALUE 0.563760 0.757712 0.577353 0.784297 HUMCAP 0.570711 0.755123 0.584688 0.783009 STRUCAP 0.487390 0.764893 0.506553 0.796545 RELCAP 0.471592 0.768069 0.492272 0.798974 LEADSH 0.702719 0.727052 0.711449 0.760140 DEVELOP 0.597104 0.742485 0.594649 0.781253 COMPENS 0.401474 0.807193 0.398755 0.814538
Cronbach's Alpha with Deleted Variable Variables
Raw Standardized