Hello,
I am currently writing my master thesis where I have a 2x2 factorial design with one downstream variable (2 independent and 2 dependent variables). I need to conduct a multivariate analysis. I conducted an online survey in Qualtrics and need to proceed with data preparation.
My problem is that each scale for each variable has several items and is repeated across four test conditions. How can I summarize the answers across all conditions to receive data for my variables, so I can use it in the multivariate analysis?
My first idea was to evaluate means per scale per condition and then put them into one column (mean response per person in a specific scale). Is it correct?
Thank you in advance!
In general, you should assess the overall reliability of each of your scales using coefficient, rather than doing so per condition, because if you used random assignment to the conditions, then you have already assumed that there will be no differences. (Also as far as I know, there is no statistical test for differences in alpha or other measures of reliability.)
Hello Olena,
I'm not fully clear on what your specific aim(s) might be. If you would explain your research question/s, what each IV is, and what each DV is (and how it is quantified), I'm sure you'd get responses that would be more focused to your needs.
Good luck with your work.
Hello Mr. Morgan, I have already assessed the overall reliability of each of the scales, thank you. My problem is rather different. I will try to explain my test below.
Thank you Mr. Morse, thank you for your answer. In an online experiment with four experimental conditions, I am testing the impact of IV personalization (high/low) under specific buying situations/ purchase decision involvement as moderator (high/low) on my DV recommendation giving capability (high/low) and DV purchase intention (high/low) in the context of shopping with voice assistants. So, I am conducting a 2x2 experimental design with one downstream variable (purchase intention). The IV personalization is evaluated on a 7-point Likert scale with 6 items, the second IV or moderator purchase decision involvement is evaluated on the 7-point Likert scale with 6 items as well. The DV recom. giving capability is assessed on a 7-point Likert scale with 6 items and the 7-point semantic differential with 4 items. The second DV/ downstream variable purchase intention is assessed on a 7-point Likert scale with 6 items.
I managed to split the four conditions in two IVs in SPSS and prepared two separate IVs for the analysis by first recoding into the same variable (IV Condition, values 1,2,3,4, e.g. 1-HighPersonalization,HighPurchaseDecInvolv in a column depending on the randomized distribution of test conditions among the participants). Then I split the conditions into two IVs by "recode into different variables" to receive two IVs with different dimensions (e.g. IV Personalisation- high or low in a column, IV PurchaseDecInvolv- high or low in a column).
How do I prepare the DVs for the analysis? I have the same Likert and semantic differential scales for assessing the DVs (recommendation giving capability and purchase intention) in every test condition. How can I summarize the answers of every item in every scale across the test conditions to conduct MANOVA? I was thinking to calculate the mean responses of the scale e.g. purchase intention per test condition (PurchaseIntent1, PurchaseIntent2...4), and then represent the results in one column under the variable name PurchaseIntentGenaral. Is it wrong?
Thank you in advance!
Hello again Olena,
Let's talk first about the DVs. Here are some options for how to treat them (each being comprised of 4 or more related items; in general, these options would apply to any Likert-type response scale having multiple items per variable):
1. Sum or average the relevant item scores together into a "total," which is then used as a DV. Rensis Likert's original conceptualization of sentiment/opinion scales was that one would use a summated score as the outcome. Of course, be sure to reverse the scoring for items posed in the opposite valence/polarity.
2. Use common factor analysis to derive linear combinations of items that appear to have both empirical association and logical association as regards your target DV(s). It can be the case that not all individual items thought to be part of identification of a single attribute/construct actually show the requisite affiliations with estimates of that construct.
3. Use Item Reponse Theory (such as the graded response model) to both derive a total score that would have interval strength and to evaluate how coherent the item set is to form a unidimensional scale.
If more than one DV needs to be considered at one time to address your specific research question(s), then the analysis should be multivariate. This could include: MANOVA, multivariate regresssion, canonical correlation, or structural equation modeling. The specific choice depends on your research question/s.
For the IVs, as you have these dichotomized, they could be handled via (m)anova, (multivariate) regression, or SEM. All can lead to potential answers about the influence of IVs that you are investigating. All can handle the moderator variable as well.
Sorry if this is a bit vague, but in some situations there are often many ways to get to a desired goal.
Good luck with your work
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS