I have some percentages and scattered themes in qualitative data. Hence, which statistical methodology will be better to categorise the qualitative data?
Qualitative data is a term used by different people to mean different things. I have a couple of statistics texts that refer to categorical data as qualitative and describe various statistical approaches for handling these (e.g., the chi square test of independence, loglinear analysis).
Also if you have percentages then you don't have (purely) qualitative data set (in the more modern sense). It is quite possible that various simple inferential statistics such as chi-square could be used depending on sample size. Whether you think that is appropriate for your data is another matter.
Statistical methods pretty much depends on the hypothesis you have set out to test. If it's just exploration, I'll go with CaRTs and categorical correspondence analysis.
Clarification on your data set structure (like what do you mean by scattered themes) will also facilitate us with better responses.
P.s. Sounds like you have a quantitative data set.
@Adam Quek I don't have quantitative dataset and the data is only qualitative. I have applied thematic analysis approach using codes. But still I am asked to apply statistical methods and I am not sure which method is most suitable if there is any.
Heh. It is still a quantitative data set if you can put together a consistent data set, eg
Subject, variable, value
A, apple, 1
A, orange, 0
B, apple, 0
B, orange, 1
where Apple and orange could be the two types of codes you used and 1/0 represent presence/absence.
In this case, exploratory tools like correspondence analysis would be great ( of course, the data set need to be in the multivariate form, ie
Subject, apple, orange
A, 1,0
B,0,1)
Just because the data is not continuous, it doesn't automatically equate as qualitative data. Categorical data can still be quantifiable.
@ Adam Quek but the coding is not associated with ranking or classes and it is associated with sub-categories or sub-headings in order to categorise the data at low level.
@Stephen, totally concur with your analogy. Echoes my very first statement actually: what is it that is being tested here?
@Imran, the examples I gave is not based on ranking too. It's based on have (which I assigned a number 1) and have not (assigned as 0). It could also be a yes or no if you're more comfortable. That is to say, the approach doesn't sound that different whether if it is a subcategory or subheadings. Once you can assign your data into category (ie nominal data) you can run any categorical based analysis.
Anyways, since you have already gotten the research done, why not provide an example of what your data actually look like? Need not be your actual data. Fake data based on your data structure you'd used would be fine.
Qualitative data is a term used by different people to mean different things. I have a couple of statistics texts that refer to categorical data as qualitative and describe various statistical approaches for handling these (e.g., the chi square test of independence, loglinear analysis).
Also if you have percentages then you don't have (purely) qualitative data set (in the more modern sense). It is quite possible that various simple inferential statistics such as chi-square could be used depending on sample size. Whether you think that is appropriate for your data is another matter.
I wonder if we are all talking about the same thing here? When I moved across from a social science department to work with medics, I found that they often talked about data from, say, a structured questionnaire survey as 'qualitative', by which they meant that it was not generated by experimental or quasi-experimental methods. However, It could still be analyzed quantitatively, using non-parametric methods as appropriate.
The key thing here is what the data are and how they were produced. If we are talking about text, generated from transcription of in-depth interviews or group discussions, or something similar, then quantitative approaches are inappropriate. If we are talking about data generated from a structured questionnaire or similar, then quantitative analysis is possible, once the data have been coded up. Hope this helps.
@Glillian here the data is generated from transcription of in-depth interviews and a questionnaire with open ended questions. Hence I also agree with you that the quantitative approaches are not suitable in my case. Thank you for you clarification please.
@Chamara your view is also same like Glillian and very informative. Hence, I am able to justify at some extent in my research the inappropriateness of quantitative methods.Thanks for your help please.
If you have a probability sample, you can apply statistical test. More likely a chi-square test. Only if you have a probability sample.
As people were mentioning above, the boundary between qual and quant starts to get fuzzy when you start counting things up. Nothing is purely one or the other.
For categorizing data, you can test for interrater reliability using Kappa if you have at least one other person check the data. Depending on your categorizations and subcategories, not too hard to check. Other things you can do is quantify coding frequencies for case comparisons. Greater numbers of categorizations in cases could tell you something about what you are looking at, so long as most people would categorize the events similarly.
Essentially, a lot of quantitative data (especially in social sciences) is qualitative data. It's been rated quantitatively in some fashion, but it is at base describing a the quality of a relationship or concept. There's no physical representation of "identity" that anyone can point to, but we can rate it's dimensionality using large scale surveys.
Likewise, observers can notice and ascribe meanings to behaviors and note the influence on others before developing an observational hypothesis on how that behavior functions in a group or community before noting it's frequency.
It's not useless to quantify qualitative data, nor do I agree that it's necessarily shearing a pig. It's simply using an inductive rather than deductive process.
Atlas-ti software gives the opportunity to generate diverse tables. These can be use as SPSS data file. This combined feature opens a whole range of quantitative analysis to be operated on qualitative data. Some are interesting for controlling your own work, but some can well be use in the writung up of a research.
Qualiative data does not rise above the ordinal level for which interquartile deviation is the appropriate statistic for interpretation.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis