Hi Edward! I see your point about confusing the participants. If you work out something from scratch, you should strive for conformity to avoid that. But as you said "established measures", I would not recommend changing the scales. If you keep them as they are, you will still have the possibility to compare with other studies using the same scales. You could work with the survey to avoid confusion, e.g., avoid lumping together questions with different metric.

In statistics you will get standardized estimates, e.g., in regression, where you can examine the relative strength of relationships. But when you interpret the estimates, it's best to relate it to the original metric of a scale, using the unstandardized estimate (e.g., a one-point gain in an independent variable resulted in a .... higher/lower score in the dependent variable).

Best

John-Kåre

Dear Edward,

it depends on the task of your survey. for the most surveys, it is the best way to have just one type of Likert scale. Hereby, you can calculate a reliability analysis with Cronbach´s Alpha or a factor analysis.

What do you want to find out with your survey?

Best regards

Tanja

Hi Tanja,

Thank you for the prompt response.

For three of the established measures, I will be looking to find the mean of the responses to each question, whilst for the fourth I will conduct a factor analysis. From this initial analysis, I will look for any realtionships between variables.

I am concerned that using three different (5,6, and 7) point Likert scales may confuse participants (adolescents).

If if I am to change the amount of points on an established likert scale measure, is there a way to work out which point is most appropriate?

Many thanks

That is possible and there are statistical tools to eventually standardize them into same scale, when you reach the stage of correlations and regression.

However, the different scales may confuse the respondents.

The terminology of the scores may change (eg 1-Very unlikely may be 1-Rare), but maintaining a 5-point would be advisable. You can then test for reliability of each contruct (group of questions) using the Cronbach alpha.

Hi Edward,

I think the best way is to have same Likert scales - I would take 5 or 7 points. I never worked with different numbers of points in a Likert scale. The exeption is, when you can analyze each Likert scale separately. Would this be an option?

Best regards

Tanja

Hi Edward! I see your point about confusing the participants. If you work out something from scratch, you should strive for conformity to avoid that. But as you said "established measures", I would not recommend changing the scales. If you keep them as they are, you will still have the possibility to compare with other studies using the same scales. You could work with the survey to avoid confusion, e.g., avoid lumping together questions with different metric.

In statistics you will get standardized estimates, e.g., in regression, where you can examine the relative strength of relationships. But when you interpret the estimates, it's best to relate it to the original metric of a scale, using the unstandardized estimate (e.g., a one-point gain in an independent variable resulted in a .... higher/lower score in the dependent variable).

Best

John-Kåre

Yes, I do agree for a uniform questionnaire, preferably 5 point, one neutral in the centre and two of a kind on either side. I personally do not go for any correlation/regression for analysis as they are not continuous variables. Just percents are fine. It is not advised to confuse the respondent keep it simple ; the question as well as the response

good luck

ravi

Dear Edward

I have conducted research with adolescents and because of their limited exposure to research instruments, using different scales might be confusing. However, keep John-Kåre Vederhus's advice on "established measures" in mind.

Kind regards

Marléne

The only occasion I used a mix of Likert scales in one questionnaire was to validate another researcher work who claimed that a move from 5-level scale to 7-level scale provides a faster way to take decisions on statements. That is, time to finish the questionnaire was a variable by itself beside the target variables studied. The questionnaire had both scales used. Results supported the claim (though different culture and country). However, to avoid difficulties, you need to be present during the administration of the questionnaires.

Dear Edward,

On most occasions using a 1 point likert scale is always more effective as it ensures Consistency & reliability to a greater extent in Research. However, there are occasions when researchers use multiple point scaled Liker scale for different dimensions/variables being studied depending on their subjectivity/objectivity.

Normally, it is the best to have a 7-point scale as it brings more freedom for respondents to give their response in the best possible manner & privides more freedom.

Best wishes,

Dear Edward,

Judging from what you seek to achieve in your study, as well as your intended target respondents, I suggest you use a uniform questionnaire.

That notwithstanding, if your questionnaires would be self-admistered, then you may utilise different-point likert scale questions as you would be present to clear any possible confusions the instrument may present to respondents. Thereafter, you would be well positioned to carry out whatever type of analysis you wish to carry out.

All the best.

The best way is to use the same Likertype scale in one survey to avoid confusion for the respondents and in statistical analysis

The multiple point scaled (Likert scale) may be used according to the objectives but uniform questionnaire would be more effective both in terms of clarity and statistical analysis.

Thanks

I totally agreed and highly recommended John-Kåre Vederhus answer

In my view according to subject of Psychological testing; lickert scale is only one that is five point rating but responses are agreed, disagreed etc. all others are rating scales, the number of responses 3, 4, 5, 6, 7, 9 depends upon the construct being measured and the depth and clarity desired.

Does anyone perhaps have paper outlining the rational for this? I am discussing a preexisting measure (that has about 3 different scales of measurement embedded within it ) which I tried to employ a bifactor model to. My understanding is that, in comparison to the other measures, the poor model fit statistics may be accounted for due to the multiple scales of measurement and the heterogeneity of the sample. It would be good to get a real understanding of this.

I think it is better to use same Likert Scale for all the questionnaires you include in your research. In one of my earlier submissions to a conference, my study was found problematic because I used 3 types of Likert Scale for 3 different questionnaire types. I guess, use of different scales create a confusion for the readers.

yes