SPSS: How to analyze and interpret Likert-Scale Questionnaire using SPSS 

Thanks you Anthony indeed it is a good one

it better you run ordinal logistic regression. it does not require criterion variable to be continuous and it does not require normal distribution assumption to be satisfied

welcome Moses

You can fit a discrete outcome model that has potentially different slopes for each category of the response thereby relaxing the proportional odds assumption. Alternatively, you could use the generalized ordered logit model.

check your statement before you write. In this example we had one independent variable ( leadership skills which by factorial reduction was decomposed into two variables ( planning and evaluation) now whether we run multiple regression or ordinal logistic regression since our dependent variable is an ordered response is based on the null - hypothesis the population originate from normal distribution. So this case I needed to see P-Value greater than 0.05. However, in this case the P-value was less than 0.05 so we went straight to do logarithm transformation. The results stilled failed the normality tests and therefore my choice was ordinal logistic regression which does not require normality assumption to be satisfied. Unless there is something you are not understanding kindly be patient to learn

my statement is on spot. i am highly trained in applied statistics and theory of estimation, so i am raising an academic issue. Kindly answer my question. is ordinal logit non parametric? Non parametric methods are distribution free while logit model is based on a cummulative distrbution function of a known distribution. Secondly is the normality envisaged about the dependent variable or about outcomes? The OLS assumption only require that the error term be identically and independently distributed. Normality assumption is about t and f tests. Your explanation is even appalling. What exactly do you mean by this 'population originate from a normal distribution' what is this normal population that is bigger than a population. You also miss the point about p-value by failing to recognize that 0.05 is not universal but rather referring to a particular level of significance. I don't doubt your knowledge of SPSS but just remember with computers GABBAGE in GABBAGE OUT. Just give statistics a chance. If you take time to watch other videos on Likert Scale you will realize that they they avoid regression. If you also take time to read OWOUR's thesis on Likert Scale and regression(British Columbia University), you'll have a different view. You can side chat me for academic engagement so that it does not look like a public spat. Otherwise you are a good presenter who should know that statistics has rules which should be followed otherwise you run into unrealistic interpretations like ' if you increase leadership by one unit' @@norwegianresearchtraininginsti

computed mean average on all implementation questions they were 24 questions in that variable implementation

we train inbox your details I will contact you

give your email to nyachome2015peter@gmail.com

i have the same question

for descriptive statistics use frequencies and bargraphs to present the data , for inferential statistics unless all assumptions of parametric tests you can run probit or logit models

@@norwegianresearchtraininginsti Thank you for responding,, was studying about Kendall's Coefficient of Concordance (to help determine level of agreement) can it work ?

@@nebby760 it can work you only make use of transformation since we require a kind of dummy variables (0) (1) where 1 is variable of interest expressing agreement

@@norwegianresearchtraininginsti Thank you,

@@norwegianresearchtraininginsti Thank you. Let me ready mote about it.

Normality tests is a precondition to run multiple linear regression or preconditions or necessary conditions ( they are all called diagnostics tests ( normality, linearity, homoscedasticity, outlier, multi-collinearity). What you are not understanding is that Likert-scale we can use to run ordinal logistic regression or Multiple linear regression if independent variables are more than two since the sum of scales if its 5 likert- scale and above can be treated as interval scale and hence qualify to run multiple linear regression if only if the subject( dependent) variable has normal distribution. This is scientific proof documented in academic literature . So absolutely there is no error in this explanation. Therefore allow me to disapprove your argument . Likert-scale can be treated as interval scale or ordinal which allow us to treat it in two different ways

I am lost for words. Linearity envisaged by OLS is of parameters not variables. This is why you can include the square of independent variables as other explanatory variables without violating any of those assumptions. Also the so called diagnostics are performed after regression not before for simple reason that most are about the error term. Econometrics gives clear ways of dealing with such violations. It is true that with limited choice dependent variable you can use logit or probit but when the categories are more than two then it is either multivariate logit or probit and the coefficients are not propensities so must be interpreted differently. for example in the case of a binary logit, the dependent variable becomes the log of odds ratio. Statistics is a science @@norwegianresearchtraininginsti

can you watch others lecturers like ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-k1jxs2xaUXI.html. One think for all analysts is that before you run any model you need to check your data if its meets all requirements for that specific model especially if you have to run parametric tests statistics. We tests for assumptions before regression not after regression since since we do some transformation if we fail. All these assumptions come in during data exploration stage.

I still maintain that all tests envisaged in OLS are done after regression because apart from that of Multicollinearity and specification, all others are about the error term. There is no way you can test for heteroscedasticity, autocorrelation without first running a regression. Take the most basic tests like Goldfeld-Quandt and Durbin-Watson respectively and you will realise that the formulation have error terms in them. For example these two tests nest on the variance/covariance matrix of the error term and you cannot get the error term before regression. You will find the following useful. 1.Kuzan, et.al ' the Seven Deadly Sins of Statistical Analysis (Annals of Plastic Surgery) 2. Dabi, M. Possible Errors in Quantitative Data Analysis ( international journal of science and technology). You can also chat me on aoberejanam@gmail.com or leave text on +254712047090 @@norwegianresearchtraininginsti

@@almadiobere2220 we can run regression since tests like autocorrelation are in regression but my point her is we do it as part of our data exploration stage before main analysis begin. What am bringing forth here is that there is need for data exploration to be done immediately you are through with data cleaning up i.e replacing missing values, eliminating outliers, variable identification etc. When we start testing for assumptions we have not yet began the analysis it is at this stage where an analyst need to use data transformation to a very great extent. When we tests normality and we fail to get normal distribution but we are interested to run parametric tests we transform by either square the variable as you said, cube it or use log to base 10. Data analysis ordinarily only take 7 days while the preparation to data a analysis takes even more than 1 month and its at this stage where proper understanding of the data is performed . so the so called diagnostics done after regression is only done to aid understanding the data. This is a mistake many do make the final stage we decide to choose multiple linear regression there is normally much we have done in the background most authors call them preliminary data analysis others sum them together as data clean up.

check this perhaps it will be better the video editing did some errrors in my voice it is english ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-A6g8NG3K_vo.html