Introduksjon til Analysemetoder Analyse av data Statistisk inferens Multivariate analysetekniker Litteratur: Churchill kap. 13-17 Troye & Grønhaug kap. 6
Chapter 13 Analyseverktøyet må tilpasses de problemstillinger utrederen ønsker å belyse There is a logical connection between defining the problem, choosing a research design, then applying the appropriate analysis techniques.
Editing (examining) the Data Chapter 13 Editing (examining) the Data This is the first step in getting to know your data. This applies to both qualitative and quantitative data. Qualitative: e.g. Is there agreement between respondents on the same phenomenon? Quantitative: e.g. Is the missing data missing at random?
Chapter 13 Coding Coding is the process of categorizing data, often through applying numbers to categories such that it can be counted. Codes should be mutually exclusive and collectively exhaustive. Multiple categories is no problem. Code book: a record of what was done. Keep in mind what you plan to do with the data.
Tabulation: counting the number of cases that fall into a category. Chapter 13 Tabulation: counting the number of cases that fall into a category. Cross-tabulation: two or more variables treated simultaneously. Cross-tabulation Men Women 8 12 20 Smokers Nonsmokers 15 9 24 23 21 44 Tabulation
Cross-tab Percentages Chapter 13 Cross-tab Percentages Cross-tabulation Men Women 18% 28% 46% Smokers Nonsmokers 34% 20% 54% 52% 48% 100% E.g. 8/44=18% Tabulation
Chapter 13 Histogram
Choice of Analysis Technique Chapter 14 Choice of Analysis Technique The appropriate technique depends on: type of data (nominal, ordinal, interval, ratio). research design. assumptions underlying the test statistic.
Statistical tests depend on certain assumptions for their validity Chapter 14 Statistical tests depend on certain assumptions for their validity t-test independent samples. normal distribution of the characteristic of interest. equal variances in two populations.
Chapter 14 Univariate Analysis Analyzing single measures of n sample objects, or multiple measures analyzed independently. e.g. You measure only sales in one or more samples, or you measure sales and attitude toward the product but don’t analyze any interaction.
Multivariate Analysis Chapter 14 Multivariate Analysis Two or more measures of n sample objects analyzed simultaneously. e.g. You measure sales and attitude toward the product and analyze the interaction.
Chapter 14 Hypothesis Testing A hypothesis can never be completely proven, there is always room for error. Therefore, hypotheses should be stated such that they are falsifiable, and such that there is an alternative hypothesis. The alternative is what the researcher argues for. Ho: Sales will not increase as a result of our advertising campaign. Ha: Sales will increase as a result of our advertising campaign.
Statistical Significance and Error Chapter 14 Statistical Significance and Error Type I error ( error): rejecting a true null hypothesis. Type II error ( error): not rejecting a false null hypothesis.
Examination of Differences Chapter 15 Examination of Differences Are the research results statistically significant, or could they have occurred by chance due to the fact that only a sample of the population was contacted? e.g. Is there a “significant” difference in sales between two groups after one group was exposed to an advertising campaign?
Chapter 15 Ethical Dilemma 15.1 The results are opposite to the hypothesized outcome. After searching the literature, the researcher finds support (a theory) that fits the results. What should they do? Rewrite the argument to fit the results? Does it mean the theory is dead?
Chapter 15 z- versus t-statistic The z-statistic assumes a known population variance, which is highly unlikely, therefore, the t-statistic is much more commonly used. They are basically two ways of accomplishing the same thing.
One-tailed test Two-tailed test Chapter 15 Ho: Sales will not increase as a result of our advertising campaign. Ha: Sales will increase as a result of our advertising campaign. Two-tailed test H0: Sales will not change as a result of our advertising campaign. Ha: Sales will change as a result of our advertising campaign.
Chapter 15 Ethical Dilemma You plan to use a t-test and have chosen a level of significance, e.g. .05. Your samples are large so the cutoff for a two-tailed test is 1.96. Your actual results (t-test) are 1.95, so you round up to 2, and are therefore faced with accepting the alternative hypothesis (what you want). Is this OK?