1
$\begingroup$

How to deal with questionnaire, where 40 questions that represent 8 independent constructs use 5-point Likert's scales and another 5 questions that represent dependent variable use 6-points Likert's scales? Do i need to standardize these last 5 questions so, they measurement scale will change to 5 - points Likert's scale? Thanks!

Clarification:

Dear All, many thanks for your guidance!

The ultimate goal of this quantitative research is to test the hypotheses: to explore if chosen IVs positively correlated with DV. There are will be a few steps as follows:

  1. Pilot study: I just collected some 50 responses to my questionnaire and I would like to perform some reliability analysis: a. Cronbach alpha; b. Convergent and Discriminant Validity tests; c. If it will be necessary – eliminate or replace some questions.
  2. Distribute questionnaire to target sample and collect the data.
  3. Analyze data: a. Repeat p. 1 + Deliberation of Goodness-of-Fit (GOF) criteria indices to evaluate overall model fit for CFA and SEM + Performing internal consistency and CFA tests; b. Structural Model Assessment and Hypotheses Testing: i. Obtaining the critical ratio (C.R./t-value): divide the regression weight estimate by the estimate of its standard error (S.E). ii. Examining paths using path estimates and CR values iii. Pearson Correlation Coefficient (r) measures the strength of relationships between independent variables and the dependent variable This would be the plan.

The English version of the questionnaire is available from under the link: https://docs.google.com/forms/d/1RPhHYXDApnnybFj1v-Il0C2N_OMs6LU9iU5MXviV5JA/viewform?c=0&w=1

With this regards, my first question will be: when we do Cronbach alpha test – do all constructs need to have the same measurement scale? Since most of the questions (40) use 5 –point Likert scale I assume, that other 5 questions (with 6 and 7-point Likert scales) also need to be transformed (standardized) into 5-poin Likert scale. Or maybe better to find Cronbach alpha separately for 40 questions that measure 8 different IV constructs and then, for 5 questions that measure DV with different measurement scales? Thanks!

$\endgroup$
1
  • $\begingroup$ It would be helpful if you provided some information on the intended statistical analysis plan. You talk about independent components so I assume you run a factor analysis or another dimension reduction technique? You also mention DVs and IVs, please add your analytical plan to make the question more specific $\endgroup$
    – StevenP
    May 12, 2015 at 14:25

2 Answers 2

4
$\begingroup$

Going just based on what you wrote, you do not need to rescale anything. I assume within each of your 8 scales, all of the items have the same response scale. Otherwise the value for your construct would be weighted more heavily by certain questions. Different response scales between your different construct scales and between your DVs isn't a problem at all. People regress response variables on criterion variables of different scales all the time; in fact it is the default for most people.

However, as StevenP stated, the answer may be different if you have specific analysis goals. But if you are wondering if your constructs significantly predict some DVs, then the F and p values will be the same standardized or not.

$\endgroup$
2
$\begingroup$

My answer to this question is related: Cronbach's alpha: dependent and independent variables

After checking the reliabilities of the 8 construct scale and the reliability of the dependent scale using Cronbach's alpha, you can just regress the total scores of the scales disregarding the underlying Likert scales.

$\endgroup$
2
  • $\begingroup$ Would it be OK to run Cronbach's alpha test separately for each construct (each is defined by 5 questions)? When I'm running test on all variables, I'm receiving a>0.7 but, many items of my questionnaire have CITC < 0.35 thus, need to be eliminated from future study. On contrary, if I calculated Cronbach's alpha separately, first for 40 items (8 IV constructs) and then, for 4 items (1DV), number of the items with CITC <0.35 is lower. When i run CrA test for each construct (5 items) independently then, I do not need to eliminate any of my items because CICT > .35 in all cases, Thanks! $\endgroup$
    – Oleg Cohan
    May 14, 2015 at 12:04
  • $\begingroup$ Yes, run Cronbach's alpha test separately for each construct. $\endgroup$
    – spdrnl
    May 14, 2015 at 23:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.