The analysis is a key and integral part of making sense of survey questions. Once you have the survey response, the next question is how to analyze the survey questionnaire. To begin with, let’s understand that there are three different categories of questions, also known as measurement scales.
The first question concerns the type of category, also known as the nominal scale. When participants select options for their answers, such as gender or ethnicity, they choose from a list of options.
Our second set of questions is on an ordinal scale. These are similar to category questions, except there is a specific sequence between the categories rather than the categories being independent.
Next, we have the interval and ratio scales, where the ratio scale is frequently used for numeric responses with absolute zero. Further, the participants might be asked an open-ended question, which needs to be coded before analysis.
Depending on question type and scale, the researcher can employ several analytical techniques to analyze the survey questionnaire.
Data cleaning
After collecting data, the first and most important task is to clean it to prepare it for the data analysis. The data cleaning exercise includes tasks like merging data sets and recoding or transforming data sets or variables.
Coding
Coding a questionnaire is another critical first step in the data cleaning and processing stage. The researcher can code for a variety of scales and questions.
One can, for instance, assign one to men and two to women based on their gender. In addition, there can be categories for age ranges, such as 15 to 20-year-olds, 21 to 25-year-olds, etc. Often, open-ended questions must be post-coded to make sense.
Analysis Plan
After the data is collected, a researcher must develop an analysis plan. This should ideally relate to my research questions and study objective.
Suppose your study is primarily quantitative. In that case, your analysis plan should align with the type of questions.
If you plan to analyze open-ended questions using word clouds, word frequencies, or sentiment analysis, include them in mixed-method research.
Descriptive Statistics and Univariate Analysis
The researcher’s first step is summarising and outlining each question’s responses. One can count the responses to categorical questions or estimate their frequency. These statistics are called “univariate” because they only look at one question or variable at a time.
An example would be researching how often females and males use a particular product. Usually, this information is presented using bar charts, pie charts, and percentages. A similar format is used to report regular inquiries, such as responses to the various age categories. One can also employ central tendency metrics to describe and summarise continuous type questions using mean or median.
Further, reporting on the dispersion or distribution of their responses is also helpful. The researcher can use a range of standard deviations as a metric. The use of standard deviations is often preferred over the use of ranges.
Bivariate Analysis
The next step is to describe the questionnaire results in more detail after performing a bivariate analysis. This entails examining question pairs to discover how they interact or diverge. Using a cross-tabulation, one can examine the relationship between two category-type questions. The age and gender of a person are two examples.
Furthermore, it might be helpful to perform a bivariate analysis using both continuous and categorical questions.
At this point, we must remember that all we are doing is summarising and describing the responses. The scatter plot provides a visual representation of the relationship between two continuous types of questions where each person’s responses are plotted.
Inferential Statistics
Inferential statistics are required if one wishes to increase the accuracy of our interpretation and provide inference about the population. Additionally, inferential statistics can be used to examine whether the results of our research were chance outcomes based on the information they provide.
If results could have been obtained randomly, they are considered insignificant. When outcomes cannot be explained by chance, they are deemed significant in the social sciences. We often put the proficiency level at 95%, which means we are 95% certain in our analysis that the result is significant.
In terms of inferential statistics, numerous inferential statistical tests exist.
A one-way chi-square test can determine whether one of the groups is statistically larger than the others. Based on my initial sample of men and women, we would inquire whether more men responded than women. An assessment of whether the difference in size between the two groups is purely coincidental would be determined by a one-way Chi-square test. We recommend the two-way chi-square test for one or more of the preliminary bivariate tests.
Further, in the case of a continuous-type question to see that the mean for men is greater than the mean for women, one employs a t-test. If our category question included more than two groups, ANOVA analysis could be used.
Assessing Relationships: Correlation and Regression
It is essential to ascertain the relationship among variables beyond bivariate analysis and inferential statistics. The best way to do that is by using correlation and regression metrics. One could use a correlation if one wanted to know whether there is a connection between these variables.
Once again, it informs whether there is a significant relationship or if it is only a coincidental one. One can also go further and assess the strength of the relationship by conducting regression analysis, wherein one can look at the change in the outcome variable due to the explanatory variable.
Kultar Singh – Chief Executive Officer, Sambodhi
