Understanding Evidence-Based Research Methods: Survey Analysis, T-Tests, and Odds Ratios
Etchegaray, Jason M., Ottenbacher, Allison J., Sittig, Dean F., McCoy, Allison B., HERD : Health Environments Research & Design Journal
This contribution to the methodology series focuses on two ways to analyze large data sets. The authors welcome comments from all readers who have suggestions about the way information is presented or questions about the content of this column. It is not comprehensive and does not replace information found in textbooks or peer-reviewed articles. Beyond their experience, the authors have used textbooks and articles as sources and recommend that you also refer to them for more detailed explanations of what is discussed in this column.
Scenario: Allan and Gwen recently finished discussing the importance of developing a data analysis plan, including checking for data usability and reliability. Now that their data set has been cleaned, they are ready to discuss some statistical analyses they can conduct. Here is their conversation.
Allan: Gwen, because our data set is quite large, I feel a bit overwhelmed and I'm not sure where to start. Any suggestions?
Gwen: Sure, Allan. Before collecting data, you told me about two main questions your boss, Jennifer, wanted us to answer with this project.
Allan: That's right! The first question she wants answered is whether those working in the perinatal intensive care unit (PICU) are as satisfied with their work environment after the PICU moved from the old hospital to the new building. What type of analysis will allow us to do that?
Gwen: Well, you have implicitly identified three key elements we need to know: the timing of our measurement, the construct of interest, and linking responses in the data set by participant. To accomplish your goal, we need to compare two main values from the data set. The first value is satisfaction level before the move; the second value is satisfaction level after the move. In this case, we measured satisfaction at both times with three questions, re ferred to in the data set as SATISFACTION1A, 2A, and 3A measured before the move and SATISFACTION1B, 2B, and 3B measured after the move. As you can see in the data set (see file located at ), the variables are located in columns and the rows each represent a participant who provided data at both time points. For example, Participant Number 1 provided scores of 6, 6, and 6 for the SATISFACTION1A, 2A, and 3A variables.
Allan: What does a score of 6 mean for these variables? I also noticed that Participant Number 1 provided scores of 7 for the SATISFACTION 1B, 2B, and 3B variables. What does that mean?
Gwen: As you will recall, we asked all participants in the PICU to answer three questions about how satisfied they were with their working conditions. For example, one of the questions was, "I am satisfied with the layout of my workspace." Each participant was instructed to answer these questions on a seven-point Likert-type scale, where 1 = Strongly Disagree and 7 = Strongly Agree. Participant Number 1 agreed with these satisfaction questions before the move and felt even more satisfied after the move, as evidenced by the increase in scores from 6 to 7.
Allan: Do we then compare SATISFACTION1A to 1B, 2A to 2B, etc.? That seems pretty straightforward.
Gwen: No, not quite. As you will recall, we decided that satisfaction was a construct you wanted to measure. We measure constructs with multiple items, and when we do so, we are able to determine whether our measurement of the construct is internally consistent. In many software packages, one can demonstrate internal consistency by computing Cronbach's alpha for construct of interest. In this case, we have two constructs: SATISFACTIONA for the three satisfaction items from the first time period; and SATISFACTIONB for the three satisfaction items from the second time period. To expedite the analysis, I used SPSS and found that Cronbach's alpha = .91 for SATISFACTIONA and .95 for SATISFACTIONB. Values above .70 are viewed as indicative of a construct being internally consistent. In this case, both constructs- SATISFACTIONA and SATISFACTIONB- are internally consistent and we can compare these constructs to each other as opposed to having to compare the individual items. …