Collected quantitative data needs to electronically organized and managed, as well. Different datasets needs to be labeled in order to keep track of all of them. Before using analyzing tools like Excel or SPSS the data needs to be checked for completeness and errors. Afterwards the statistical tools can be used to generate data representations such as frequency tables,
correlations, regressions, distribution graphics or independence tests to describe the data and make patterns among variables visible.
For our purposes, we intend to use the five-number summary and standard deviation, to ana-lyze the basic statistical properties from our quantitative data, which we will collect during Delphi round two and online surveying. The five-number summary consists of the max, min, mean, quarter percentiles, and standard deviation values for each question or sub-question.
3.11.1 Hypothesis testing
Hypothesis testing is based on a tentative assumption about a population parameter, which can be based on the normal distribution, or not, and is used to test validity of a claim or re-search hypothesis (Anderson, Sweeney, Williams, Freeman, & Shoesmith, 2009, s. 291), where the tentative assumption is referred to as the null hypothesis, denoted by H0, and the opposite hypothesis to the null hypothesis is called the alternative hypothesis, denoted by H1
(ibid p288). The H1 hypothesis is the one we want to reach, and to be proved, by rejecting the H0 hypothesis (ibid p290). The probability of making a Type I error when the null hypothesis is true as an equality is referred to as the level of significance, which is denoted by α, and if the cost of making a Type I error is high, then a small value of α is preferred (Anderson, Sweeney, Williams, Freeman, & Shoesmith, 2009, s. 294), however, a smaller value will in-crease the Type II errors (ibid p328), thus selecting a proper level of significance determines the power of the test, and thus the probability of correctly rejecting the H0 when it is false.
3.11.2 Sampling
For our usage of the Delphi method we intend to use a non-probabilistic judgement sampling, which would never the less limit potential non-sampling errors due to lack of knowledge (Anderson, Sweeney, Williams, Freeman, & Shoesmith, 2009), adhering to the requirement of the Delphi method to ensure that the respondents are SMEs, with knowledge both about server virtualization, but also knowledgeable about company and organizations reasoning for choosing and deploying server virtualization.
Due to the nature and requirements of applying a Delphi method, such as questioning subject matter experts in their field of expertise, we anticipate that for this method, due to the allotted timeframe, we might not receive a samples size, which will allow us to statistically analyze the data and reach our desired certainty of 95%. However, nevertheless we will apply the methods, allowing for the possibility of uncertainty, and correlate the statistical findings we do achieve, with those of Delphi first round, and Interview series.
For our Online Survey method we also intend to use a non-probabilistic judgement sampling, which would also in this case limit potential non-sampling errors due to lack of knowledge (Anderson, Sweeney, Williams, Freeman, & Shoesmith, 2009), since we would require selection assistance from participating Promoter companies to provide us with potential respondents, who all should have knowledge about server virtualization, but also
knowledgeable about their company and organizations reasoning for choosing and deploying server virtualization.
The responding sample size for our Online Survey, should preferably superseded 30, thus given a response rate of 25% we would have to distribute the survey to at least 120 potential respondents. The general form of the interval estimate of the population mean µ being x±Margin of Error (Anderson, Sweeney, Williams, Freeman, & Shoesmith, 2009, s. 260), denoted by ∆, which can be calculated using za/2(σ/√n), thus x± za/2(σ/√n), depending on the confidence level, where 99% gives za/2=2.576 and 95% (our choice) gives za/2=1.960, depend-ing on that the population follows a normal distribution in which case the calculated confi-dence interval will be exact, otherwise it will be approximate (ibid).
3.11.3 Validity and Reliability
Our approach to achieve statistical reliability and validity, are based in our attempt to em-brace the scientific concepts of rigor, thoroughness, openness to alternative or rival conclu-sions and tests, and adherence to measuring protocols, and especially for quantitative meth-odology also thorough sampling, and determining appropriate statistical tests and methods to correlate data, would also be appropriate, if the measured data and the test methods, are ap-propriate to be able to evaluate and draw correlated conclusions from.
The purpose of quantitative validity is to reduce, or eliminate, error and bias in the variables, and for the test of tests, to ensure high correlation between the data, tests and criteria. Valid-ity evolves around the concept of ensuring that the measurements, tests, and criterion are in-deed measuring, testing for which the criterion is set. Validity in this respect can be viewed in two ways for our purposes. The first involve validity of the test itself and is the degree to which evidence and theory support the interpretation of the intended tests to test for what was intended to test for.
Reliability on the other hand is concerned with the consistency of the measurement and that measures are not taken in error or with bias, thus allowing for erroneous or biased data to be evaluated (whether for sample or population). Thus the reliability determines how repeated measures, by the same collection methods or techniques, render the same measurements and thus evaluation. So with proper sampling methods, the sample results will provide “good”
estimates of the population parameters (Anderson, Sweeney, Williams, Freeman, &
Shoesmith, 2009, s. 224), and there are certain methods that can be used to evaluate statistical reliability, such as the Cronbach’s alfa (Wikipedia contributors, 2009).
4 Results & Analysis
As mentioned in chapter 1 Introduction & Problem Area, our primary research question was to determine if Data Centre power consumption reduction is related to, or perceived as, an important (success) factor for implementing and deploying server virtualization for
consolidation purposes, and if not, what other decision areas affect server virtualization and power consumption reduction, respectively.
In this chapter we describe the results and analysis for each method we employed, Delphi (qualitatively and quantitatively), and Interviews (qualitatively).