Chi Square for Likert Scale in Single Sample

#1
Evening All,

I do hope someone can shed some light on my fog of the chi square. I conducted a single sample survey using Likert Scale scored 1-5 (1=Strongly Agree to 5=Strongly Disagree) to survey employees in medical practices on leadership behaviors during a major transition. The goal was to determine if leader behaviors influenced employee morale and attitudes. IV - Leader behaviors DVs- employee morale & employee attitudes.

The following are the three stated hypotheses:

First Hypotheses

H0: Leadership does not affect employee attitudes during the transition from paper medical records to EMR in medical groups.

H1: Leadership does affect employee attitudes during the transition from paper medical records to EMR in medical groups.

Second Hypothesis:

H0: Leadership does not affect employee morale during the transition from paper medical records to EMR in medical groups.

H1: Leadership does affect employee morale during the transition from paper medical records to EMR in medical groups.

Third Hypothesis:

H0: There is no correlation between employee attitudes and employee morale in medical groups during the transition from paper medical records to EMR.

H1: There is a positive correlation between employee attitudes and employee morale in medical groups during the transition from paper medical records to EMR.

The sample size was 55 employees working in front and back office positions in medical offices. The study used cluster random sampling. 1st section of the survey first asked education level, years in healthcare industry, years at current employer, and what area (front office or back office) did the employee work. The rest of the survey was Likert 1-5 with 20 questions.
Sample of questions:

My practice manager encouraged teamwork and cooperation during the transition.

I was able to express my concerns without fear during the transition.

My practice manager offered positive reinforcement when employees seemed stressed during the transition

The dissertation chair and I did agreed on using chi square with cross tabulation to test if there are existing relationships between specific pairs of questions. That part I can comprehend.

Where I am lost is that he felt that chi square test for independence would be the best analysis to test the 3 stated hypotheses. Everything I am reading is not making sense. What I want to know is how, using the 20 likert scale questions and having one sample, would I test the hypotheses using the chi square test for independence in order to reject or retain the null hypotheses? I'm looking at my responses as scale data.

I did read something that said to combine the two spectrums of answers to narrow the field down to 3. In other words combine all strongly agree with somewhat agree and strongly disagree with somewhat disagree to have 3 responses of strongly agree, neutral, and strongly disagree however I'm afraid that was not revealed in my approved proposal and that it will be opening a bigger can of worms!
I am using SPSS for data analysis.

I'm a total rookie at statistics. So I am hoping someone would be kind enough to shed some light.
 

CowboyBear

Super Moderator
#2
Could you provide more information about how you are attempting to measure each of your constructs (e.g., leadership, employee attitudes, employee morale)? Are each of these constructs assessed using multiple Likert items, or just one each? If there are multiple items per construct, looking at cross-tabs of pairs of questions may not be relevant. Instead you may need to sum coded responses across multiple items to get a score for each participant on each construct, and then proceed with a different analysis.

The chi square test may not be appropriate here regardless. The chi square test is usually used when you're looking at the relationship between two nominal variables. But you have data that is (at least) ordinal. The chi square may be able to tell you if there is a relationship of some kind between variables, but not the direction of the relationship. Something like Spearman's rho correlation might be more useful to you (or parametric correlation or regression, if you are willing to treat the Likert data as continuous).

Collapsing the Likert items into smaller categories does not make sense unless you have a very clear reason for doing so. This kind of thing bins information and reduces statistical power.
 

CowboyBear

Super Moderator
#3
Also... you are probably aware of this, but it sounds like you are doing purely survey research. Which is not really suitable for drawing causal conclusions, at least not without applying additional strategies to support such conclusions. So the hypotheses you're testing (that refer to leadership affecting things) may not be testable with the data you have. You will be able to look at relationships between variables, though.