recently I have collected data( survey) and I want to analyze the data so for that I am going to prepare my data, I need your useful suggestion that what should be the first step after data collection ( for example: normality , reliability , correlation, and or other) and what test should be use for normality or others??
need useful suggestions of the experts
Arsalan -
You haven't said what your goals are, or what kind of data you have.
The first step is not data collection. The first step is to plan what you want to learn; what data you need to collect to do that; what sample design, data collection methodology and estimation or prediction method you need to accomplish what you want to do, to discover what you need to learn. Then you collect your data in a manner appropriate to the plan you have. If you collect data and then want to know what to do with it, you may not have collected what you need to do much useful with it at all. This is a very common mistake.
Doing this backward, collect the data and then planning, you are going to have unnecessary limitations. But since it may be too late to do this properly, perhaps you can do something. So ... you need to explain how you collected your data, and what you want to do with it. If you did not use a probability-of-selection-based approach or a model-based approach (so, with regressor data), or collect a census, then you will not likely have very much that you can do. If that is the case, you will have to be cautious not to overstate the usefulness of your results.
Best wishes - Jim
James R Knaub thank you for you response to my question. first of all i am collecting data for my thesis, which is based on questionnaire and data were collected on random sampling technique, now I am going to check the missing values and same answers on one item to remove those all.
but I want to know more good techniques the expert using such as normality, alpha and so on.
Is there some reason you are interested on normality? It is generally not needed. It can be helpful for easy confidence intervals, but if you want a confidence interval around a mean, the Central Limit Theorem helps. Also, Chebyshev's inequality may be helpful. You can show a relative standard error. There is a lot of information available on other distributions.
For prediction (i.e., regression), it is generally desirable for the estimated residuals (or better yet, the random factors of the estimated residuals for weighted least squares regression) to be approximately normally distributed, but the data for the dependent and independent variables can be distributed in any manner. Normality is not necessary. I worked a great deal with highly skewed data used in regressions.
As for alpha, I expect you mean a p-value, but a lone p-value is meaningless. You need a sensitivity analysis such as a power analysis or an effect size study. If you can use a confidence interval or just a relative standard error, that will tell you much more than a p-value, and as much as a complete hypothesis test for purposes of decision making.
I suggest that you start with some graphics. Scatterplots and other graphs can often tell you more than anything else. What graphics you might want to use depends upon what you are trying to show. Using scatterplots to show relationships between different data can be very useful.
You said "...now I am going to check the missing values and same answers on one item to remove those all." I'm not sure what you mean, but if you have missing data, it makes a big difference if you have ignorable or nonignorable missing data. If you are trying to estimate totals, then ignorable is not really "ignorable," it just means that the mechanism for generating the data is the same for all data in the area of the missing data. You might impute with mean values and it would not bias results, but it would artificially lower the variance. However, you could adjust survey weights instead of imputing.
If you are not looking for totals then I guess maybe you could "ignore" such cases. You just have a smaller random sample. But if there was a reason for missing data that make them special, so the process for generating these data is nonignorable, then you have more to consider. If you know similar data, you might randomly chose an imputed value from that. Another method might be regression. You could research "nonignorable nonresponse."
James R Knaub Mr James that's really nice explanation. usually after paper submission they ask about missing data, normalit, skewness' etc that's why I want to clean my data before starting analysis. and I was trying to normalized my dependent and independent variables for regression and path analysis. where as you said that could be in any manner and now I am not going to do that. while I will use graphics to show the relationships.
thank you Sir
I don't know anything about path analysis. You'd better check out the requirements. - Best wishes.
I think you have to look into your objective and according to that you have to arrange your data in an Excel sheet and according to your objective decide appropriate data analysis tests. Even you can use Excel it's self for data analysis or you can use SPSS software for analysis.
The main things are you have to fulfill your objective by your data analysis plan.
Best wishes
Hemal Patel
There are no hard and fast rules with regards to the next she after data collection.
What you do with the data collected is entirely up to the researcher.
What is the topic of your thesis?
What are your goals? I mean, what do you intend to achieve?
What kind of data do you have? Large or small samples?
With regards to the missing data, what have you decided? Do you want to adjust surge weight or ignore missing data?.
Your answers could go a long way in helping us to help you.
Donalben Onome Eke thank you for nice comments, Dr I know all but I ave seen most of the researchers they first normalize the data whether they are testing correlation or regression, but here I am going to check the relation and the regression where using path analysis, so for these I am removing missing data and make them normalize and my sample is more than 2000
Hemal M Patel thank you sir. actually I am using R for my data analysis and with heavy demographic information
- Badges
- Science topic