Organizations use surveys to find solutions in complex situations or seek information of a general nature such as how to prepare marketing plans, identifying the cause and symptoms of employee discontent in large factories, and more.
A primate concern of surveys is data integrity. Surveys collect information from many respondents. The data collected rarely fits the criteria of a random sample, and resultant variability poses the risk of the analysis generating a distorted view of the population. The integrity of the survey data depends on the survey design considering:
- Sampling weighs, or the inverse of the probability of being included in the sample due to the sampling design. High weighs increase the chance of distortions.
- Primary sampling unit, or the first unit sampled for design. The integrity of the survey data depends on the accuracy of such sample.
- Strata, or breaking up the population into different groups using variables such as gender, race or other factors.
- Margin of error, or the probability of discrepancy owing to sample size. A low margin of error reduces the chances of distortions.
Methods of Analysis
Surveys usually seek quantitative data, and as such, the common methods of data analysis for quantitative data finds use to analyze survey data. Some of the common methods include:
- Statistical processes such as T tests, correlation analysis, regression analysis, standard errors, significance tests, ANOVA, and scatter plots.
- Ratio estimates, estimates of odds ratio, hazard ratio estimates.
- Estimates of population means and totals, population proportions.
- Domain analysis, or analyzing systems to find common and variable parts.
- Logical analysis by applying flow charts, comparison tables, charts and diagrams, and more.
To illustrate an analysis of a survey, consider the hypothetical case of a company undertaking a market survey to understand the spending pattern of households, as a precursor to launch a financial planning service.
The population or the universe of such a study is all families with at least one income-earning member. The survey obviously cannot cover all income earners in the United States, and as such, administers the questionnaire to a sample group. The challenge before the researcher is to ensure the selected sample represents the universe, and for this, the researcher applies weights, margin of error of the sample size, and other techniques. The researcher may also undertake a test survey with one identified primary sampling unit to test the effectiveness of the sample, and the validity and soundness of the questionnaire and analysis methodology.
Before analyzing the collected data, the researcher needs to sort the data based on respondents, or stratify the sample. The common basis of categorization is according to demographics or income levels, age, geographical location, education, and more. Separating the sample into such categories allows the researcher to identify trends. For instance, people in a particular age group may tend to spend more on entertainment, people with children may tend to save a greater proportion of their income, people in a particular geographical area may tend to demonstrate some unique spending patters, and more.
The analysis of data depends on the survey design. Here, the analysis may include:
- T-test to validate hypothesis based on assumptions such as “people with two children below five years of age have a children’s education plan.”
- Correlation analysis to link savings and age.
- ANOVA to analyze the variance of spending on certain categories such as food, entertainment, and lifestyle by age and geography, and more.
World Bank Study
Consider the study to determine factors contributing to a successful community-based co-management of coastal resources among Pacific Island countries, conducted by the World Bank in 2002 for a real-life example of a survey analysis. The survey spreads across 31 sites in 5 countries, and involved 133 interviews with mini-focus groups of two to six respondents from different households. The study selected Fiji, Palau, Samoa, Solomon Islands and Tonga as representative countries for the range of coastal management conditions under study. The 31 selected sites covered the range of conditions perceived to influence management success. The basis of such perception was collecting trends for factors such as perceived catch per unit effort (CPUE), condition of habitats, threats to the site, and assessment of compliance. The focus group was selected using non-probabilistic sampling.
The methods of data collection involved:
- A five-point scale questionnaire administered on the sample.
- Referring to records with the fisheries and environmental ministries in each country and at site level.
- Recording descriptive perceptions from the focus groups.
The information resided at four levels: country, site, focus group and specific resource, habitat, threat or rule. The method of data analysis was multi-level modeling, or generalization of linear models, where data varies at multiple levels. Preparing a multi-level model requires selecting random and fixed variables. Here, the objective was to identify the perception of CPUE trend by averaging the perception scores across the three resources. The country effect was the fixed effect, and the sites and focus groups the random effects.
- UCLA Academy Technology Service. “Introduction to Survey Data Analysis.” https://www.ats.ucla.edu/stat/seminars/svy_intro/default.htm. Retrieved June 11, 2011.
- Hall, John, F. “Introduction to Survey Analysis.” https://www.marketresearchworld.net/index.php?option=com_content&task=view&id=544&Itemid=60. Retrieved June 11, 2011.
- “Sample Survey Design and Analysis.” https://support.sas.com/rnd/app/da/new/dasurvey.html. Retreived June 11, 2011.
- Chromy, James R. & Abeyasekera, Savitri. “Statistical analysis of survey data.” https://unstats.un.org/unsd/hhsurveys/FinalPublication/ch19fin3.pdf. Retrieved June 11, 2011.
Image Credit: freedigitalphotos.net/Sujin Jetkasettakorn