If it is qualitative, why would you use a scale? Tt is a contradiction.

Dear Audil,

firstly, the choice of the right scale always depends a bit on what you later on want to do with the data and what kind of questions you like to ask to whom.

However, I must confess that I am a bit confused ... in qualitative research you rarely collect quantitative data ... so what is the scale for? Usually, you do qualitative research if you do not know enough of a research context to formulate a a hypothesis. As soon as you can formulate a clear hypothesis, you go on with quantitative research to prove/decline the hypothesis - and here, you will need scales. Or did I maybe misunderstand you?

Related to your stated topic "information on socio-economic impacts", do you already know if there are impacts or don't you? If you know there are impacts and can classify them (you just want to know about the deviation), you rather should do quantitative research. If you have no idea on what such impacts may be (the reason why you cannot define a questionnaire), the question of a scale is obsolete: Than you rather do structured/non structured interviews or maybe a Delphi study if you have a chance to get hand close enough on users (in case of Delphi, they at least need to contribute twice...). Another opportunity would be action research as design. But in non of those scenarios, you need to think about a scale (in the second set in the Delphi you need to but here, usually it is a rating scale, Likert type)..

All my best,

Thomas

I think you have to consider qualitative techniques. In social studies, focus groups, interviews or semi structured are convenient.

Russ Bernard in Research Methods in Anthropology lists a number of scales used in research that combines qualitative and quantitative approaches. I agree with the authors above about scales and qualitative research, but identifying economic information about informants using a scale may be helpful in interpretation. In Brazil, national criteria are used on a five-point scale that uses education of household head and household possessions and staff (A through E). You mention impact, though, and normally we associate measures of impact with quantitative outcomes. Qualitative outcomes might be if someone feels their economic situation has improved, and if they think it's due to the project...

One of the most widely used packages to analyse qualitative information (text based) is Nvivo. As soon as you have a scale in mind you seem to be deciding for a quantitative analysis. Perhaps you could clarify what you have in mind?

Thanks everyone for your valuable response on the query. I think I need to refine my question a bit.

Actually I am working on assessing the socio-economic impact of Financial Inclusion (or access to Finance) on beneficiaries. The study consists of both the economic and social variables. For economic variables I am using a structured questionnaire in which respondent would be asked to respond to a quantitative variable like Income (Pre and Post) etc. However, the study also consists of social variables like effect on decision making or effect on participation in discussions etc. For those social variables I am planning to use a 5-point Likert Scale.

My question actually was pertaining to using a 5-Point Likert scale for analysis rather than just data collection. After collecting the data on the above mentioned scale, the analysis has to be made. My question actually is if I would use 5-point scale would it be worthwhile for arriving at a proper analysis. I mean how feasible is it to use 5-point for analysis in SPSS or other Statistical Softwares. And also if there is some other scale which can be more helpful when analysing the data and using various (not simplistic rather) statistical techniques.

Looking forward for your educated response.

With all due deference to the greater experience and expertise of the other contributers, I wonder if we're talking about the same thing. If by qualitative evaluation we mean evaluation that uses subjective input then it is possible to quantify it. For example, the rate of failure in a particular course would be clearly quantitative. But if we asked students after the first week about their confidence of being able to pass the course, that might be considered a quantifiable qualitative matter, or do I have my terminology completely wrong? Perhaps the question originally posed depends on terminology and the questioner's understanding of "qualitative." As someone with little formal preparation in this area, I would appreciate it if those with greater expertise and experience could clarify for me how well decided the definition of qualitative evaluation is and if the line between qualitative and quantitative evaluation is hard and fast (and where is it?).

I think the line between qualitative and quantitative moves, depending on the type of social science study. In perception based barometer studies, perceptions (qualitative) are converted into quantitative data. This is one of the reasons why social science is considered a soft science. Randomised sampling does not generate the same result every time, although in theory it should. It differs in relation to time and events and of course we have to consider Heisenberg's effect (that subjects under observation temporarily alter their behaviour and give different answers to what they would normally). For example, conducting a perception based survey on public perceptions state corruption by means of a quantitative barometer study just after a major governance scandal will potentially yield different results if done at a time when state corruption has not been reported in the media for some time. If the respondents are under the impression that the survey is conducted by the government by government officials in a country in a state of governance distress, they are likely to change their responses due to a fear of being traced.

I usually try to learn by observing how others do it - in the case of Audil,s research, I would look to Gallup World Poll methodology and to Afrobarometer to frame the quant component. In Quals, I would certainly look at participatory action research methods using semi-structured questionnaires and key informant interviews that support the data, a triangulation of a special sort. If three data collection methods deliver the same data, chances are the data has a higher degree of reliability. Research design in social science is always challenging and very interesting.

Hi Audil, refering to your last post, since you are planning to use VAS/Likert scales, you are on the quantitative path. And yes, it is perfectly fine to use them (the VAS scales) and analyse them with the common statistical techniques you mention. If I were doing this from the design stage, I would choose 7 levels (instead of 5) in my scales. Regards, Nora

Thanks Nora Thanks Justin,

Margarita: But is it sufficient enough for statistical analysis. I mean Agree Disagree.

Dear Audil,

the point with the Likert scale is the fact that you produce ordinal scaled data. Actually, as far as I see in the social sciences' literature, the Likert scale seems to be the most used scale in this context. However, according to Stevens, the available methodologies you can use for interpretation is rather small: Ordinal scaled data do not provide any information on the actual distance between the single scale values. In a 5-point Likert scale, the distance between 2 and 3, e.g., is not necessarily the same as between 4 and 5. Also, you do not have an absolute zero-point (not sure if this term is correct in English). Thus, if you follow the rules of statistical hard-liners, you can only calculate sums, percentages, median, quantiles and quartiles; that's all (often this anyways is enough). Truly, today there are some multivariate opportunities for ordinal scaled values, too but classical multivariate methodologies might not be possible to apply (factor analysis, cluster analysis, ...). On the other hand, there is a broad discussion which contradicts the opinion of hardliners like Stevens, and the protagonists say surely it is possible to apply (higher) multivariate statistics as long as you check that the results are sound. Just for a brief outline, the idea behind this discussion is that ordinal scaled data often are more than just ordinal scaled data. When you e.g., take the mean of examination marks, the result perfectly makes sense although taking the mean on such examination marks (which are ordinal scaled) is not appropriate. A first suitable check is calculating mean and standard deviation and contrasting the results against the median and quartiles (for my own investigation I used the 40 and 60 quartile). (in my own research I experienced Likert-based datasets that were perfectly sound regarding higher statistics and others where already the standard deviation seemed more than a "bit strange" - this at least shows that a careful check is absolutely relevant. However, as for me I realized that the median is an excellent and reliable measure and often, multivariate statistics are done but rarely needed (to answer a certain research question). In case you really need multivariate statistics and want to use the Likert scale, you can manipulate the Likert scale a bit and thus, raise the reliability of your results regarding higher statistics, e.g. raise the number of answer-options (not urgently needed but helpful) and just name the minimum/maximum (so that the participants "feel" a constant difference between those poles). Latest when you start using ordinal scaled data in SPS (you mentioned that), you may need to binaries your data (sum up positive and negative answers). But you may ask again when this happens ...

Whatever you finally choose to do, you will need to argument why you think it is appropriate. A good start into the 50 years lasting and very controversy discussion (I experienced this as being very exciting) is the article of Knapp, which I mentioned before: (Knapp, T.R. (1989). Treating Ordinal Scales as Interval Scales: An Attempt To Solve The Controversy. Nursing Research, 39(2), pp. 121-123.). However, this article is not what you can use as a single standing argument to justify why you do something rather forbidden (in the economics discipline, concepts which are accepted in Social Science are not necessarily also accepted. many economists are statisticians and thus hardliners. Here, often it is difficult to argument qualitative research ... but "abusing" ordinal scaled data ... huh...). The article of Stevens (hardliner pro distinguished-use-of-methods) I mentioned before you can find at: (Stevens, S.S. (1946). On the Theory of Scales of Measurement. Science, New Series, 103(2684), pp. 677-680.). The contradicting position originally (at least I did not found an earlier critique) was represented by Lord (Lord, F.M. (1953). On the statistical treatment of football numbers. American Psychologist, 8(12), pp. 750-751.). For the beginning this might be a good start and the overview, Knapp gives on related literature helps a lot. The discussion still is going on, thus there are much younger resources, which you additionally should use for argumentation.

Regarding the 5-point Likert scale you additionally would have to argument (particularly in a phd thesis) why you use an unpaired scale and provide a zero point ("I have no idea what I should say...") option. Why don't you use a paired number of answer options and force people to take a position? Also here, a lot of significant papers have been written, which finally are easy to find using google search,

What I finally still do not understand is what all this (particularly your second post) has to do with qualitative research - which was the original topic. Further, in your second post you use the term "structured questionnaire". I never have heard of an unstructured questionnaire as technical term ... however, unfortunately I see too many seemingly unstructured questionnaires. I guess you meant a structured interview (interviews can indeed be structured, semi-structured, or unstructured)? Please pay attention not to confuse those terms.

All my best and good luck with your research. If you need information on literature in this field, be free to contact me directly. I went through exactly those topics about two years ago and thus, the whole "thing" still is quite fresh.

yours, Thomas

Btw, the 2-Point Likert scale (agree/not agree), which Margarita suggested is exactly what is meant above with binarizing. Just with binarizing later on, you have a "full" Likert Scale available and reduce it later to those two poles: The advantage against just implementing two poles from the beginning is that you have both, a more detailed distinguishing AND the opportunity to do higher statistics.

Use of different kind of indices, descriptive statistics, or in case you would like to use econometric tools, then dummy variables (avoiding dummy trap), logit/probit model, panel data analysis may prove to be beneficial when both quantitative and qualitative variables are simultaneously used.

You should refrain from using "scales" as such in qualitative research since you are almost always tempted to go "quantitative." Try to stay away from "scales" if you truly want to be "qualitative." Just, please, use unstructured interviews with open-ended questions related to your research study.

I think Anjan Chakrabarti has a good point.

Yes, I agree with Anjan. I think we can change the qualitative data into dummies (which behaves as the quantitative data) while analyzing the data which has both quantitative and qualitative variables. Or we can bin the quantitative data to change it like categorical data.

I agree with Muammer Cetingok. Refrain from using scales if you want to be qualitative.