Mathematically, there is no formal basis for any use of any quantitative method for Likert-scale data. Empirically, countless studies have demonstrated how invaluable and accurate numerous different kinds of statistical analyses of Likert-scale data can be. The problem lies mainly in linguistics. The scale itself is robust in most ways because whatever lexical constructions used (e.g., "strongly disagree..."strongly agree" or "Very X...Very not X"), the evaluations specific to that ordering of that scale. However, this is also a problem. First, any and every response set to some questionnaire with Likert-scale questions is specific to those responses/that questionnaire. Second, all kinds of judgment biases (from variations in interpretations of the questions to "hedging", "clustering", and similar response biases).

All that is obviously of no use to you. So here's an attempt to rectify that:

1) Generally speaking, statistical methods that rely on variance as a function of mean deviance are susceptible to multiple problems: violations of non-normality, violations of homoscedasticity, and one significant advantage of these tests (robustness against some violations of assumptions) tend to result in type I errors.

2) Virtually all the most frequent statistical methods (z-scores, t-tests, ANOVA, etc.) assume not only quantitative data but continuity. While many discrete datasets (e.g., IQ scores or income distributions) approximate continuity because of the large variability possible with large intervals, likert-scales are usually 5-point scales. Measures that assume continuously distributed variables can pose problems particularly if the number of question sets is small and the participant pool is small. The more sets of responses you can get, the more you can approximate continuity.

3) There are a fair number of methods designed specifically for this type of data (e.g., Item response theory, categorical data analyses, multidimensional nonlinear descriptive analysis, classification analyses/algorithms, fuzzy probability & statistics, and dissimilarity/similarity analyses. Luckily, even if one isn't familiar with any of these, there are some fairly simple distance measures you can use to compare individual responses for each participant to better understand whether their answers are all over the place or tend to fall into groups (the latter is the preferred result). Basically, if you get a bunch of answers and most of your participants tend to respond to each question in one or two ways or tend to respond in total in ways that fall into several similar groupings, then this alone tells you something about the target population's conceptions regarding your questions.

4) Could you clarify the "10 dependent variables" = "10 hypotheses" statement? Because it could mean a serious multicollinearity problem or be inconsequential depending upon what you mean.

5) Quite apart from your questionnaire question types, number of participants, how your sample participants were recruited, etc., there is the question of what your research question is. Analyzing variances is suited to some question types but not others, and as you are employing such an analysis the reason for doing so as well as what it is intended to tell you and how it relates to other data gathered also factor into how you might wish to approach the Likert-scale portions.

I realize that these are all very general answers but I have an unfortunate tendency to avoid generalizations like the plague. However, if others don't provide far superior responses, I'd be more than happy to supply more comprehensive answers if you could supply more detail.

My sister is a nurse in a level 3 (?) neo-natal intensive care unit. I asked her about whether she had received any training, courses, seminars, etc., in emergency response procedures other than the most general, and she had not. Are you sampling from a specific subpopulation of nurses?

Mathematically, there is no formal basis for any use of any quantitative method for Likert-scale data. Empirically, countless studies have demonstrated how invaluable and accurate numerous different kinds of statistical analyses of Likert-scale data can be. The problem lies mainly in linguistics. The scale itself is robust in most ways because whatever lexical constructions used (e.g., "strongly disagree..."strongly agree" or "Very X...Very not X"), the evaluations specific to that ordering of that scale. However, this is also a problem. First, any and every response set to some questionnaire with Likert-scale questions is specific to those responses/that questionnaire. Second, all kinds of judgment biases (from variations in interpretations of the questions to "hedging", "clustering", and similar response biases).

All that is obviously of no use to you. So here's an attempt to rectify that:

1) Generally speaking, statistical methods that rely on variance as a function of mean deviance are susceptible to multiple problems: violations of non-normality, violations of homoscedasticity, and one significant advantage of these tests (robustness against some violations of assumptions) tend to result in type I errors.

2) Virtually all the most frequent statistical methods (z-scores, t-tests, ANOVA, etc.) assume not only quantitative data but continuity. While many discrete datasets (e.g., IQ scores or income distributions) approximate continuity because of the large variability possible with large intervals, likert-scales are usually 5-point scales. Measures that assume continuously distributed variables can pose problems particularly if the number of question sets is small and the participant pool is small. The more sets of responses you can get, the more you can approximate continuity.

3) There are a fair number of methods designed specifically for this type of data (e.g., Item response theory, categorical data analyses, multidimensional nonlinear descriptive analysis, classification analyses/algorithms, fuzzy probability & statistics, and dissimilarity/similarity analyses. Luckily, even if one isn't familiar with any of these, there are some fairly simple distance measures you can use to compare individual responses for each participant to better understand whether their answers are all over the place or tend to fall into groups (the latter is the preferred result). Basically, if you get a bunch of answers and most of your participants tend to respond to each question in one or two ways or tend to respond in total in ways that fall into several similar groupings, then this alone tells you something about the target population's conceptions regarding your questions.

4) Could you clarify the "10 dependent variables" = "10 hypotheses" statement? Because it could mean a serious multicollinearity problem or be inconsequential depending upon what you mean.

5) Quite apart from your questionnaire question types, number of participants, how your sample participants were recruited, etc., there is the question of what your research question is. Analyzing variances is suited to some question types but not others, and as you are employing such an analysis the reason for doing so as well as what it is intended to tell you and how it relates to other data gathered also factor into how you might wish to approach the Likert-scale portions.

I realize that these are all very general answers but I have an unfortunate tendency to avoid generalizations like the plague. However, if others don't provide far superior responses, I'd be more than happy to supply more comprehensive answers if you could supply more detail.

My sister is a nurse in a level 3 (?) neo-natal intensive care unit. I asked her about whether she had received any training, courses, seminars, etc., in emergency response procedures other than the most general, and she had not. Are you sampling from a specific subpopulation of nurses?

My goodness that is a lot of very foreign information! To jump to your last question, I am recruiting nurses from four major Little Rock hospitals...the only inclusion criteria is that they are employed by the hospital. And just to step back a minute, I have zero experience in statistics...somehow managed to go through my BSM, MSM, and PhD programs with no stats classes, which looking back, has not benefited me at all. I cannot really share openly in this forum specifics about my questionnaire because I am still awaiting Research Plan Approval so I do not want this 'out there' prematurely. My biggest hurdle at this point is answering the section on my Scientific Merit Review document is about how I plan to analyze my data. I have had my questions reviewed by several people that have concluded the questions should adequately answer my research questions, so that is not a big concern at this point. I am just trying to get a grasp, not even looking for a firm grasp, on the basics of types of measurements and how to approach each one in the simplest, most effective way.

Sorry about that. I have a range of research interests and what must approach a clinical incapacity to simplify, summarize, or provide answers that do not end up ranging all over the place. However, surprisingly one of the most powerful methods is the simplest: graph your data. Create a scatter plot of responses, a bar graph, calculate ALL measures of central tendency (mean, median, mode), and then just look at your response sets (both between each participant's responses and between all participant's responses for each question). This informal, intuitive, and almost wholly non-quantitative approach tells you more about your data than any single statistic and won't give you some single value ostensibly telling you how "significant" your results are based on assumptions that make this value worthless.

If you find your responses to questions are all over the place, this tells you something. If you find that responses tend to cluster to the middle, this tells you something.

There's no one right answer to analyzing data, which is why looking at it in various ways before picking a test can be so vital. For example, imagine that 90% of participants who answer question 4 with response "mostly disagree" (or whatever) also respond to questions 6, 8 & 14 the same way as well as never providing response y to questions 2, 7, 12, 13, & 15, this indicates that certain responses have predictive power (given a participant's response to question N, it is highly likely that they answered some set of other questions in similar ways). This may not give you a nice value, but it can be even more informative and can tell you whether some statistical test is probably very useful. Of course, one can't know whether a test is useful if one isn't familiar with the tests.

So as I consider a better way to respond to your question of tests, perhaps these will be of some use to give you a background that requires minimal familiarity with statistics (especially if you've had a decent exposure to other math courses such as a semester or 3 of calculus).

Analyzing and Interpreting Data From Likert-Type Scales

(http://www.jgme.org/doi/pdf/10.4300/JGME-5-4-18)

Generation of an Interval Metric Scale to Measure Attitude

(http://sgo.sagepub.com/content/4/1/2158244013516768)

Nurses' attitudes to a medical emergency team service in a teaching hospital

(http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2464889/pdf/427.pdf)

Frasure, J. (2008). Analysis of instruments measuring nurses’ attitudes towards research utilization: a systematic review. Journal of Advanced Nursing, 61(1), 5-18.

After much consideration, soul-searching, 3 or 4 moral crises, several meditative retreats into isolated deserts, mountains, etc., and consultation with spiritual advisors from around the world (which is pretty impressive for a day's work), I decided that perhaps the best way to answer your question (given how much I don't know about the nature of our scale, participant number, etc., among other potential issues) would be to provide groups of links to sources categorized by type/kind (from actual examples of Likert analyses in self-efficacy studies to general, free reference material) so that you could navigate them more easily. That way I provide more links than you need, but in a way that allows you to immediately exclude useless ones, check different explanations in reference materials easily as needed, and in general have more than you need but without having to sort through it.

The Basics of the Basics:

Likert Scale

(http://www.al-huda.net/2012/PA/2014/topic-dane-likert.pdf#page=2)

Analyzing Likert Data

(http://www.joe.org/joe/2012april/pdf/JOE_v50_2tt2.pdf)

Gathering Data and Assessing Results: Data Analysis

(http://nnlm.gov/evaluation/guide/stage5.pdf#page=7)

Statistics Glossary

(http://www.stats.gla.ac.uk/steps/glossary/alphabet.html)

Statistics Explained: Glossary of statistical terms

(http://www.animatedsoftware.com/elearning/Statistics%20Explained/glossary/se_glossary.html)

Concise intermediate sources on Likert-type data analysis:

(for terms, references to statistical tests, etc., that you don't recognize, see below for links to general e-textbooks/reference material and the glossaries above)

Estimating ordinal reliability for Likert-type and ordinal item response data: A conceptual, empirical, and practical guide

(http://pareonline.net/pdf/v17n3.pdf)

Powerful and Consistent Analysis of Likert-Type Rating Scales

(http://dmrussell.net/CHI2010/docs/p2391.pdf)

Likert Scales: How to (ab)use them

(http://xa.yimg.com/kq/groups/18751725/128169439/name/1LikertScales.pdf)

An item response theory analysis of self-report measures of adult attachment

(http://www.web-research-design.net/PDF/FW%26B2000.pdf)

[this paper is excellent for a few reasons: 1) Although it is a research paper, the researchers dedicate a section to the basics of Item Response Theory (ITR), a powerful, robust tool for analyzing likert-type data 2) It doesn't concern self-efficacy, but it is a paper on analyzing actual, collected Likert-type psychological/trait/attitude self-report data of different types and comparing analysis methods & 3) it is among the better examples of Likert-type data analysis out there]

The Fuzziness of Verbal Response Scales: The STAI-T Questionnaire

(http://www.atlantis-press.com/php/download_paper.php?id=2221)

[Fuzzy set theory is a fairly intuitive way to analyze linguistic data; however, most of the papers and sources I know of assume both a knowledge of statistics AND fuzzy set theory; a pretty good exception is the dissertation "A New Likert Scale Based on Fuzzy Sets Theory" by Cheryl Qing Li. However, I couldn't find a link to it that didn't require access to ProQuest. There's nothing wrong with the above paper, it's just not as introductory as I'd like]

Longer, intermediate level sources on Likert-type analysis:

An Introduction to Modern Measurement Theory

(http://appliedresearch.cancer.gov/archive/cognitive/immt.pdf)

Measurement

(http://psych.colorado.edu/~mcclella/measchap.pdf)

The Basics of Item Response Theory

(http://echo.edres.org:8080/irt/baker/final.pdf)

A "non‐technical overview of multidimensional scaling" paper from Tutorials in Quantitative Methods in Psychology

(http://www.tqmp.org/Content/vol05-1/p001/p001.pdf)

Likert-type data analysis examples specific to self-efficacy measures:

Developing and validating a tool to measure parenting self-efficacy

The SCI Exercise Self-Efficacy Scale (ESES): development and psychometric properties

(http://www.ijbnpa.org/content/pdf/1479-5868-4-34.pdf)

The Nursing Competence Self-Efficacy Scale (NCSES): An Instrument Development and Psychometric Assessment Study

(http://dalspace.library.dal.ca:8080/bitstream/handle/10222/27969/Kennedy-Evelyn-PhD-Nursing-%20June-2013.pdf.pdf?sequence=1)

see also http://echo.edres.org:8080/irt/

Creating and Validating Self-Efficacy Scales for Students

(http://usir.salford.ac.uk/17673/4/Creating_and_Validating_Self-Efficacy_Scales_for_Students.pdf)

Linking transformational leadership to nurses’ extra-role performance: the mediating role of self-efficacy and work engagement

(http://www.researchgate.net/publication/229056425_Linking_transformational_leadership_to_nurses'_extrarole_performance_the_mediating_role_of_selfefficacy_and_work_engagement/file/79e41508ff53dc2aa4.pdf)

General e-textbooks/reference material to use for looking things up found in sources listed above:

Statistics at Square One (9th Ed.)

(http://www.bmj.com/about-bmj/resources-readers/publications/statistics-square-one)

Surfstat.australia: an online text in introductory Statistics

(http://surfstat.anu.edu.au/surfstat-home/contents.html)

The Little Handbook of Statistical Practice

(http://www.jerrydallal.com/LHSP/LHSP.htm)

Concepts & Applications of Inferential Statistics

(http://vassarstats.net/textbook/)

Probability and statistics EBook

(http://wiki.stat.ucla.edu/socr/index.php/EBook)

Article Linking transformational leadership to nurses' extra-role pe...

Short answer, use R Program, download from Internet, fairly straightforward!

Another somewhat short answer:

1. In the best psych journals, summative scales -- of what Likert and multiple choice exams are special cases -- are evaluated by GLM analyses (eg, ANOVAs and regression).

2. Stats are logical models. If they give reasonable results, then they are OK. Point 1 above shows they give reasonable results (which is, it seems, why the multiple references noted are seldom used in stats classes).

3. A general purpose stats program I have designed from the ground up for the multivariate GLM (of which ANOVA, contingency table chi squares, and regression are special cases) is admittedly designed for those wishing to use the full power of the GLM, but it is turning out to be very easy to use. UniMult.com.

4. Absolutely essential that you know what the meaning of correlation and confidence intervals, and that you have your work reviewed by someone who knows the meaning of the terms in point 3.

1) "The best psych journals" have repeatedly published studies that either the same or similar journals have critiqued not specifically but in general for regular, systematic, even demanded errors in the underlying logic behind statistical models and methods. I've referred to some of that literature here before.

2) ANOVA is not "multivariate", hence "MANOVA".

Point well made re the "best journals". I should have said that I have published successfully in the best journals using 5-7 item response format. One of the reasons I failed to mention for these type items working is that we do not analyze the item but rather add together a number of them so the data analyzed is more continuous. When I get some time, I'm looking forward to examining the literature you quote and thinking on why they work OK given the problems that have been raised.

I used "ANVOA" only because it is more easily understood by someone with almost no stats. In my teaching and data analysis I use the multivariate GLM instead of ANOVA and MANOVA excepted as happens when they describe a special case I happen to run. A true multivariate GLM allows mixing of nominal and other variables among both IVs and DVs, and interactions across nominal and other variables (including curvilinear elements). With this model, terms such as ANOVA are primarily jargon needed to read the literature.

Sorry, I submitted before fully answering -- use R program, Fischer's Exact Test if non-parametric data for Likert-scales.