I can see total attitude score as interval, therefore, ANOVA with demographics. Further, if you  use counts then chi-square 

Thanks for the suggestions

You may want to upvote my suggestions. Nevertheless, I think those 2 tests are what your analysis needs. 

In the situation you describe I see only rank order data and no interval scale that meet S.S. Steven's criteria. https://en.wikipedia.org/wiki/Level_of_measurement. So the Spearman rank order correlation coefficient would appear to be appropriate statistic to use in answering your questions IF your data are monotonic in distribution. https://statistics.laerd.com/statistical-guides/spearmans-rank-order-correlation-statistical-guide.php.

Are you treating each category (e.g., Money) as a variable that could be ranked 1 ,2, or 3? Or are you treating the Rank #1 as a variable with different categories, according to which topic was selected?

Also is this an independent or a dependent variable?

Whatever test you use, you need to be aware that a p value does not tell you what most people think it does--the probability that your results are due to chance. The failure of "statistically significant" results to replicate has become a major scandal in psychology and medicine. See attached paper.

Agree whole-heartedly with the spearman correlation for your two questions. And you can't anova the demographics because your dependent variable (the financial priorities) isn't continuous. You could use the anova on the scale question. A chi-square is a possibility with the financial priorities x settlement size or other demographics but if you would need to ensure every cell had >5 examples and if your distribution is very skewed (i.e. mostly big settlement sizes) then you're going to have a statistically sig chi-square but not telling you much. Plot the preferences against settlement size and see what that looks like in the context of the spearman. I realize that your rank list 1,2,3 etc is categorical data, so you're looking at a line with text on it basically, but sometimes that can tell an interesting story. for example, as your settlement size is larger, cash is less important than credit, or something like that.

Also agree with Joseph about the value. the nice thing about spearman is that it doubles as an effect size measurement.

Good luck!

While spearman rank correlation may be a descriptive step, it should not be the study. There are many ways to analyze results from a questionnaire. I noted that chi-square is a possibility and am happy that Andrews agrees with me.

I can see attitude scale scores and demographics as an ANOVA. Attitude score is interval and we can ANOVA/ANCOVA - -to make a stronger statement, we can do a regression of Attitudes scores being conditioned on demographics. Regressors do not have to be continuous. BTW, are binary variables continuous?  

Sub-analyses can be undertaken:  a sub-analysis of those who those who thought money was a priority coded as 1 and zero otherwise can be regressed on demographics; another regression of those who choose financial security coded as ! and zero otherwise can be regressed on demographics. Consequently, there are many ways of dragging the animal home. Simply performing a rank correlation is not a study. Suppose, we take all who choose money and those who didn't (coded zero)  and logit that with Attitude score plus demographics and did same with financial security and those who didn't as zeros logit- regressed on Attitude score and demographics appropriately coded. Hidden information are not discovered by correlation coefficient. BTW never use  Wikipedia in academic work.     

I agree that a logit for each possible answer provides more detail for each answer choice but doesn't appear to be what was asked and doesn't rank the importance of the options against each other. A logit would work if respondents could "select all that apply", and allow you to relate each financial item back to the dependent variable. There's no priority information with that, however. Just because more people selected "money" than selected "credit" doesn't mean that money is more important to people. Since the author collected priority information (sounds like the items were ranked by respondents), a spearman provides information about if the importance of the items with respect to each other correlates with the demographics or scale or whatever dependent variable. If you did want to run a logit, you could run it for just the top priority for each person. i.e. 1 for if "money" was selected as top priority, 0 if it wasn't. in a different column code 1 if "credit" was the #1 priority, 0 if it wasn't, etc. (I'm making up these columns, but you get the idea.) That's one way to run a logit with your currently collected data. I submit that the logit and spearman answer two different questions. The spearman makes sense if there is some theoretical sense of rank (such as money in hand vs credit vs collateral).

and agreed that you can run an anova/ancova with demographics as independent variables against your scale. but I wouldn't do that with your ranking system as the dependent variable. No, regressions don't have to be continuous dependent variables. They also don't have to be binary. You can run logistic regressions with multiple dependent categories, but we're talking really high level stuff now. I'm guessing by the original question that we're not going there.

and while I also don't endorse wiki for formal citations, it is correct in this case :-)

What explains the importance of an option is the important research issue and not simply spearman correlation which are used for descriptive  analysis.

"In my questionaire I asked respondents to name the three most important priorities (in order of importance) in connection with a specific area of life."               

(a) Money   (b)Financial Security (c) Home

For example, they have the above--  Andrew, you are saying we cannot provide logit or Tobit or Probit for those who selected money coded as 1 and zero, otherwise as I discussed earlier.

Similarly, Fin. Security and the other treated the same way. We agree on chi-square,

but we do not agree on this (please read my sub-analyses well because I know we can unless am not reading you comments well (am  in a movie theatre)

Then, you don't see that my sub-analyses provide more detailed research than just telling one that x and Y correlate via spearman rank?  

Chuck,

I hope you are not depending on any single analytic approach. I usually try many alternative approaches to see what differences they disclose.

Sorry for citing wiki but I wanted to short-cut going to the library. Were you to submit a manuscript for publication, I would look for evidence that your operationalized concepts or constructs met APA-defined reliability and validity criteria . . . AND provided the relevant journal citations.

You have many options in going forward. Remember, however, that the burden of proof of concept and procedure is on the author of any claimed research "findings."

We await the results of your analysis.

Hi John:

Please, I hope you realize that that the wiki comment was not directed to you. I simply wanted to indicate that wiki is not used in academic work. Maybe, I should have said it in a different way, but I honestly wasn't directing my remarks to you. I get along with persons named "John" and I will appreciate your overlooking my framing of what I said. The truth is that it was not directed to you. If you think that it was--I apologize.

.

Chuck,

No offense taken. In blog sites like this I try to find relevant information that is instantaneously accessible. The important thing is to gather all inputs relating to your inquiry and make judicious choices, knowing the inevitable strengths and weaknesses among them.

Best wishes on successful pursuit of your research.

Attila, to provide clarity, since we've run into methodological circles:

I think Chuck's logit idea is fine. And we've both now agreed on using the ANOVA/ANCOVA for the scale data.

He and I seem to be interpreting what you've collected differently, however, with you narrative responses. If you simply coded the three important items, in any order, so you now have a single column of 1's, 2's, 3's etc corresponding to each option, then sure, I would agree the logit can be applied to the whole thing by recoding each option as a 1-0, basically dummy variables. If, however, you have 3 different columns for 1st, 2nd, and 3rd priority, each column coded with 1's, 2's, 3's for each option, then I would offer that your best option, if you want to perform the logit, is to do it just the the 1st priority column. Otherwise it gets messy.

If you have ranks assigned to your different financial options, i.e. most popular to least popular, I see no problem looking at a spearman, and for your own purposes, plotting it. Doing this has helped me in the past see some qualitative relationships that I didn't see before. The plot is for your own brainstorming as you consider the data. The spearman shows how the importance of one item, relative to the others, may predict something else. That's my piece, and I leave it to you and your chair or mentors to discuss further if Chuck disagrees with my logic. We'll just have to respectfully agree to disagree.

It just goes to show there's more than one way to look at data. 

Good luck! I hope all of this somehow helped in the end.

Hi John;

Thank you; I deeply appreciate your kindness. I have many friends named"John" and they are all wonderful, generous and extremely kind. Thank God, you're not offended. This happened because I was in a movie theatre was dictating to my phone and wasn't paying attention. I promise you, going forward, I will be much more careful. Take care and kind wishes!!  

 Andrew:

Thank you for your comments. I truly have enjoyed the interaction, that is, why I love academics and research. You are a '"kool" guy and you handle it very well. Best wishes to you, Andrew!  

This seems to be a complex problem, and it reminds me of some very useful advice I received many years ago in graduate school: Design your data collection according to your analysis plan. In other words, decide in advance how you will analyze the data that you are considering collecting.

Likewise, Chuck. I enjoyed it, too.

David: couldn't agree more!

All,

The discussion about choice of appropriate statistical method for given data metrics didn't mention whether the data were derived from a probability sample from a known universe. I assumed they were but, if not, the inferential statistics under discussion would not be applicable . . . except "heuristically," i.e., to guide development of more rigorous future research.

Dear All

Thanks you for your suggestions. Andrew, I do have my priorities in 3 different columns. I sampled mayors counrywide, but as I had no financial resources I could not make it a probability sampling, also it was anonymous.

Hi, Attila,

You appear to have exploratory data that should be analyzed in many ways. "Ranked nominal" is probably not a useful way of looking at the coded priorities. I suggest a Chi square contingency table to determine if there are any significant group differences using your demographic data to create groups. Phi coefficient could then be used to determine the magnitude of the relationship.

If you determine that  your ranked nominal data is actually better understood to be ordinal, then group comparisons can be made with the Mann Whitney U test. Spearman rho would be the correct correlation to run to compare your attitudinal scale (if interval/ratio) with ordinal data.

The more interesting test, from my perspective, would be to recode your priorities into dummy sets (1-Money, 0-other) and run a logistic regression to identify significant predictors of Money as a priority. Other priorities that you are interested in could also be tested this way. This may require considerable recoding of predictor variables to satisfy the assumptions of logistic regression.

Attila:

Peter Kindle's response has some useful suggestions for analyzing your data. Unfortunately, NO statistical significance test can answer the questions you raise--"Is there a connection between certain priorities" and "Are respondents who scored low on certain variables of the attitude scale more likely to have certain prioroties." All such tests can tell you is the probability, IF there were no correlation between these variables in the population sampled, that you would find the correlation you did, or a larger one, by chance sampling variation alone. The idea that such tests tell you the proabiliity that your results are actually due to chance is a long-standing and widespread misunderstanding of such tests, one that has repeatedly been challenged by prominent statisticians, and it has played a major role in the recent "crisis of replicability" in psychology and medicine. For a more detailed explanation, see the attached paper, pp. 16-18, and the references cited there.

This kind of data is very difficult to analyze because the categories aren't really separate (i.e., what is your first priority might be my second).

As Peter Kindle says, one possibility would be to recode your data as 1 or 0 (mentioned or not). You could do this to recode each topic mentioned as first priority or not. A similar option would be to code whether each topic was mentioned or not across all three priorities. This would still give you quite a few dichotomous variables to analyze.

I agree with the suggestions made by Peter. But recoding in 1-0 will be useful if you are looking to analyze that way. If the categories are more than two, then this needs to be separately analysed for each pair using the tests proposed and then synthesize the results.

Dear Peter

Thanks a lot for your suggestions. I created dummy variables and ran regression.

Attila