My main issue that im not able to find suitable analysis methods rather than descriptive analysis, i have found the percentage of each question and connect it to the research i have done then i have also find the standard dev, mode, median and so on. But my prof is asking me for high level statistics but am not able to apply any high level statistics on my current data.

one more question, using chi square test, how do we calculate the expected value ?!

Nora, the data analysis method and techniques should help you answer your research question. In order to help you, we need to know more about your data (e.g, your hypotheses, dependent and independent variable/s) and the research aims and objectives. 

Nora, your response variables ("days taking to return": a-d, "use of social media": yes/no/don't care etc) are ordinal variables. Please do not calculate things like mean and standard deviation (nothing where a sum has to be calculated, because sums of such values are not defined). You can use order statistics like mode, median, interquertile range and so on.

When your prof is asking you for "high level statistics" then please ask your prof what exactly he means with this. Ask him what his precise research questions are, what his precise hypotheses are, and ask him what methods he thinks you should apply to answer these questions. This is the responsibility of your prof to tell you and to guide you. Instist on getting this help.

The expected values for a chi-squared test are obtained from the marginals of the frequency tables. There are 1001 sites you will fnd when you google for "chi-squared test expected values".

Nora, are you a bachelors or Masters student. In either case you should have undertaken (or had instruction) in basic parametric and non-parametric statistical tests, if your tutor/supervisor expects some analysis? Also normally one wouldn't seek answers to questions via a survey unless you had a very good idea of the hypotheses you were to tests. You design the questions to address your research questions and/or objectives?

The data you have described can be treated as nominal data for which you can undertake chi-square tests of association/independence, for example testing the null hypothesis that there is no association between the use of social media for Education purposes and the ways of communication etc....

If you have more variables of the type you have listed then (treating them all as nominal variables) you could subject them to a multiple correspondence analysis (MCA) and obtain a segmentation map.

I do not consider chi-square high level statistics whereas MCA is a multivariate analysis.

For Bachelors (final year dissertation) the sort of analysis my social science department would expect from a survey would be parametric tests such as t-tests, ANOVA, correlations (on individual or collapsed variables), regression and maybe multiple regression (obviously not all are necessary but this presents the level) and non-parametric tests such as chi-square, rank correlation, MCA

For Masters theses the sort of analysis my social science department would expect from a survey would be parametric tests such as t-tests, ANOVA, correlations (on individual or collapsed variables), regression, multiple regression, factor analysis and possibly some structural equation modelling (obviously not all are necessary but this presents the level) and non-parametric tests such as chi-square, rank correlation, MCA, cluster analysis

I should point out these requirements are usually underpinned by a research methods course (in many cases where a student has prepared a proposal for dissertation or thesis outlining aims, objectives, methodology and method, design of data collection instrument(s) and has a clear idea of what tests they will undertake - if quantitative survey).

Dear Nora,

I am not sure what your tutor means but I think what he means (and what I would ask if I you were my student) is an analysis that combines several questions to reach a conclusion on research question.  I suppose you have more questions than what you posted, but lets give an example from what you posted.  Obviously interaction between students and staff change if a different communication medium is used.  One aspect of that is response time.

Unfortunately you cannot cleanly answer this question because you have not asked about students and staff separately in your first question. None the less you can compare the answers on the first question between the two groups "Social media for Educational purpose users" and "Social media for educational purposes non users" 

This is the simple idea and just comparing means and standard deviations  or probably better median's and quartiles should already give you a reasonably good idea (i.e. if the difference between the median's is a sizable fraction of the difference between the median and the quartiles, you maybe on to something).  To really find out you have to do a statistical analysis e.g by doing a Kruskal Wallis test to see if there is a significant difference.

Caveat!: Significant is not the same as relevant: if you have an insane number of survey results. Minute differences maybe "significant" which just means you can detect your survey tot behave from random. Which is almost certain to happen if you have enough surveys. 

Caveat: if you play this game enough time then, most likely you will find a "significant" result. That is to be expected: if say alpha = 5%, the chance of finding a "significant result" by chance can be as high as   ( 1 - (1-5%)^20) = about 2/3

Dear Nora,

The only suggestion that I can give is: which analysis method is OK depends on not only which kind of data you have but also what you want to know. So, if you describe your research questions clearly, more precise suggestions may be offered.

Nora, 

I use SPSS for statistical analysis.  There is a great resource that not only provides coaching for the software but also very detailed, decision-tree formatted counsel on statistics.  It is https://statistics.laerd.com/ and I have found it well-worth the modest subscription fee charged for its use.

https://statistics.laerd.com/

Nora

the expected values for sets of tabulated data (cross tabs) can be determined as follows: add the totals for each row and each column. Then determine the overall totals for colums and rows (theses should be equal). For each individual observation in the cells of a table there is a unique row total and column total for that cell. The expected value of any observation in the table is given by (row total * column total)/(overall total) or (RT*CT)/OT the result will be that an expected value for each cell can be determined. The sum of the row expected values should equal the sum of column expected totals. The chi-square statistic is determined by subtracting each expected value (E) from its corresponding observed value (O) and squaring each of the (O-E)^2 values dividing by E (avoiding zero sum issue). The products are summed therefore chi-square statistic = Sum (O-E)^2/E

Hi Nora,

Generally, it is proper practice to determine your research questions prior to data collection. This helps to avoid “data snooping” which is a practice that generally leads to inflated false positive rates and at best is appropriate only in a “pilot” study scenario where you will follow up with additional testing on those things that you think you’ve identified from the pilot. It also ensures that you ask the right questions and don’t have to repeat the survey.

1. For any individual question, provided you have a large enough sample (think hundreds) you could use a confidence interval to estimate the proportion who answer one specific answer of your choice (or some combination). For example you might estimate the percentage of (students?) who get answers same/next day by combining categories.

2. If you want to look at cross-tabs, a chi-square test for independence (any two questions) could be appropriate, assuming that sample size conditions are met.

3. Nora, you might want to consider rewording some of the questions. In particular question 2 should have re-worded responses and question 3 should just be a “select all that apply”.

I hope this helps you. Best wishes.