I have dependent and independent variables in either 5 or 7 point scales, strongly disagree to strongly agree, not important at all to extremely important etc. One usual way is take mean scores and apply ANOVA and/or Regression depending upon hypothesis or research question.
I also read an article where 1-5 scales were assigned weights 1 to 5 and total score for each item was calcuted and then for the construct.
I am curious if there are any other way to anslyse scaled data?
It will be really helpfull if some literature could be referred to in this regard.
Best regards,
Ali
The standard practice is to calculate the sum (or average) of the scores for the items that go into a scale. However, you also need to assess whether the items do indeed form a reliable score. Since there have been so many questions about this topic here, I have assembled some resources on it:
https://www.researchgate.net/post/How_can_one_help_people_create_simple_scales_from_Likert-scored_items
Hello,
Just small addition. 5 to 7 answer categories (Likert type) maybe treated as either categorical or continuous and this is an important decision while running analysis (in Mplus u apply different estimators - ML for continuous and WLSMV for categorical).
Regards, Witold
Thanks Witold, for introducing another tool (MPlus). I will look into it. :)
If you are testing theory , you can use Structural Equation Modelling.
you may refer to the following book;
Multivariate Data Analysis (7th Edition)
by Joseph F. Hair Jr (Author), William C. Black (Author), Barry J. Babin (Author), Rolph E. Anderson
Thanks Parathasarathy. Actually I am trying to validate a taxonomy and a model which was earlier suggested based of qualitative data. Now I am trying to validate it with help of bigger sample.
7 point sacle may be considered as sacle data rather than categorical buy it depends on the purpose of your analysis I hink. If you want to run a model like regression and you going to take the variable which has 7 point scale as dependent variable, it is better to use ordered probit regression model.
Thanks Ramanayaka, your response have valuable suggestion. I will look into ordered probit regression model. Thanks.
Dear Ali, You may use factor analysis. Thanks.
You may also read the book entitled "Categorical Data Analysis, Third Edition by ALAN AGRESTI".
First of all , it is imperative to establish the construct validity of your instrument and you may compute factor scores.
As for analysis , you may use descriptive statistics but it is recommended that based on review of literature you have already formulated hypotheses to be tested and you have built a hypothetical model to be tested using Path analysis, Elaboration of variables or other multivariate techniques
You should rely more on Kruskal–Wallis test by ranks, since your variables are measured in an ordinal scale. for example http://users.sussex.ac.uk/~grahamh/RM1web/Kruskal-Wallis%20Handoout2011.pdf
The problem with Kruskal–Wallis is that you have to analyze your original variables one at a time, rather than combining them into a single, more reliable scale.
Thanks for the heads-up David.
I am in data sorting and cleaning phase and thus not sure as yet if the data fulfills condition of normality or otherwise. However, when I went through Anduela's suggested link and further searching that "each condition is performed by a different group of participants; i.e. you have an independent-measures design with three or more conditions. " or We also assume that the cases of each group are independent. I have following situation:
1. IV is 21 Internet usage patterns collected on 5 point Likert scale. Based upon higher score for each pattern, the 21 patterns are combined into 7 thematic patterns each containing 1 to 5 patterns. Same group of participants have recorded their scores for all the patterns. Does it mean my data does not fulfill the condition for Kruskal-Wallis test as 7 thematic patterns (conditions/groups) are from same group of participants?
2. My DV is Importance of Information Security Threats, which is recorded for 75 different threats [one Likert Type item for each threat on 7 points(1 for not at all important to 7: Extremely important). For unsure/don't know cases 0 will used. It is not possible to combine them and thus be dealt as 75 independent variables [number seems too big though].
At the moment, I am planning to divide IV into 7 groups and then analyze density plot for 75 threats. In this way, I would find 5 most and 5 least important threats for each group, based upon Mean Score and SD. Even I can find participants with certain Internet usage patterns who are not aware of specific threat(s).
Just curious if there could be other ways to connect Internet usage patterns with Threats' importance and awareness?
[N.B. David, if my answer is available in the link you have shared in this thread for the very first time, .I am still going through your suggested thread but haven't gone through it completely as yet :) ]
- Badges
- Science topic