I have a sample of 363 participants and 6 quantitative Likert scale variables in a correlational design. I am exploring the relationship between tertiary education students' mental health literacy, current mental health, help-seeking intentions, self-stigma, public-stigma, and self-reliance. I would also like to explore whether or not gender has an effect on the relationships. What is the best way to analyse this data?
I have been thinking of a regression analysis then a mediation analysis, or structural equation modelling? Alternatively, I have also been considering ordinal logistic regression.
What are your research questions? Without knowing difficult to recommend a procedure, except maybe a scatter plot matrix. Are you interested in how much each is associated with each other? How they are associated as a group? If the six tap into a single dimension? Also worth knowing what their histograms are like.
Thanks for your response Sarah Brock
. For the pairwise associations, the scatter plot matrices will be important. Depending what the univariate distributions are like and if the scatter plots suggests something similar to a linear relationship, some type of correlation will allow provide a numeric estimate of this. As far as combining the measures, the choice will also depend on the descriptive statistics. There are lots of papers on using scree plots to help decide the number of dimensions (e.g., doi = 10.1037/met0000200). If you are thinking of mediation models, you should have a more detailed research question specifying what may mediate what. Regarding gender, you could see if it is associated with either the dimensions (e.g., the factors from an EFA) or the individual items (then adjusting p values). Trying to examine if gender influences these is more tricky, and some people think you cannot show (see doi 10.2307/2289064).
It seems that your research is exploratory in nature. If so, you could consider partial least squares structural equation modeling (PLS-SEM), which is ideally suited for explorative research purposes (Hair et al., 2011; Hair et al., 2022). Furthermore, PLS-SEM has been recommended over regression for mediation analysis (Ramli et al., 2018). Here are the full citations.
Hair, J. F., Hult, G. T. M., Ringle, C. M., & Sarstedt, M. (2022). A primer on partial least squares structural equation modeling (PLS-SEM) (3rd ed.). SAGE Publications. https://us.sagepub.com/en-us/nam/a-primer-on-partial-least-squares-structural-equation-modeling-pls-sem/book270548
Hair, J. F., Ringle, C. M., & Sarstedt, M. (2011). PLS-SEM: Indeed a silver bullet. Journal of Marketing Theory and Practice, 19(2), 139–152. https://doi.org/10.2753/MTP1069-6679190202
Ramli, N. A., Latan, H., & Nartea, G. V. (2018). Why should PLS-SEM be used rather than regression? Evidence from the capital structure perspective. In N. K. Avkiran & C. M. Ringle (Eds.), Partial Least Squares Structural Equation Modeling: Recent Advances in Banking and Finance (pp. 171–209). Springer International Publishing. https://doi.org/10.1007/978-3-319-71691-6_6
Good luck,
...you can test using Pearson if your data is parametric (ratio/interval) and meeting the normality assumption or Spearman/Kendall if your data is non parametric, and fail to met the above conditions.
You may use independent t-tests to test significance differences based on gender, or one way ANOVA depending on the nature of your data.
Please note, that irrespective of research questions, you must take into account the fact that you have ordered categorical data - the numerical data indicate only an ordering, not a numerical quantitative value with standard size and distance. Only non-parametric rank-based methods for description and -if needed- for analyses are appropriate. Svensson E. Guidelines to statistical evaluations of data from rating scales and questionnaires. JRehab Med 2001;33:47-48 and other papers by me.
I have analysed the effect of gender towards people with mental health problems. You find this paper in my research: (PDF) Impact of gender and age on attitudes towards mental illness in Sweden.html
Sarah Brock
, just quietly, it's Likert, not Linkert. And in case it needs mentioning, the first syllable rhymes with "lick" not "like". : -)All the best with your research.
Sarah: Perhaps, you may try "ANOVA 2-factor with replication" and see the p-value for testing the Hypothesis 'if there is interaction between your 2 factors of interest' ... technically, the latter is the Alternative, while the Hypothesis is H_0: "Factors A and B do not interact to affect the mean responses". However, if p-value is small then you may say that you have enough evidence to reject H_0 in favor of the Alternative H_A: "Factors A and B do interact to affect the mean responses" - sorry, for this long writing.
a similar one from 2015 with 29 answers in discussion:
https://www.researchgate.net/post/Am-I-able-to-use-ANOVA-to-analyse-likert-scale-data
Sarah, for some reason I can't see all the answers here, so I apologise about dupication. I agree with the need to use non-parametric tests.
If you wish to use single indicator items (especially for the dependent variables), then this is now very approachable using bootstrapping. See the excellent textbook by Andrew Hayes. http://afhayes.com/introduction-to-mediation-moderation-and-conditional-process-analysis.html One thing he opened my eyes to was how mediation and moderation differ and sometimes when you have a dichotomous variable like gender, the distinction is important.
If you have multiple indicator variables and wish to use structural equation modelling, consider robust weighted least squares estimator in MPLUS, for categorical variables. All the best
- Badges
- Science topic
- Similar topics
- Biological Science
- Biochemistry
- Biochemical Techniques
- Analytical Chemistry
- pH