Help on Dissertation Statistics?

Discussion in 'General Distance Learning Discussions' started by JeepNerd, Nov 27, 2015.

Loading...
  1. JeepNerd

    JeepNerd New Member

    I started by searching the forums for statistics and I am seeing various threads about getting degrees in this area, but not a lot of threads outside of that.

    I believe this will be an appropriate thread of discussion for those who are studying at a distance but will need help on their dissertation with statistics?

    What I am hoping to find is some resources, perhaps some of them internal (other DI participants) and suggestions on what to read, who to hire when you are working at distance!

    Those that may not have seen me posting before, I am working on my DBA and I am currently writing my research proposal. I got the response back from the committee and they believe the research is valid, but I am using incorrect terms and suggesting in appropriate tests (z test) for my sample.

    What I am looking for is resources that will allow me to explain my concept and help me correctly identify an appropriate design. If you believe it would be appropriate I could post up some details here, but I am not sure if we want to clog up the forums with those types of threads?

    So, are there other forums or places you suggest that I go to? Should I hire one of these online dissertation / statistics consultants? Any personal experience with these consultants?

    I know my research needs to be my own and want to disclose everything, so do I put this in my follow up to my proposal? That I hired XYZ as a consultant to further my education and selection of the appropriate design and tests?

    Some of the consultants also offer their services for the research itself and that makes sense in that they can make sure I am not making inappropriate conclusions on the results of those calculations?

    Books? Papers..?

    Thanks in advance!
    Sam
     
  2. Kizmet

    Kizmet Moderator

    I've never been in your position so I have no first hand experience in these matters but it is my understanding that hiring a consultant(s) to help you through the stats and/or edit your dis is not uncommon.
     
  3. Rich Douglas

    Rich Douglas Well-Known Member

    Hiring a consultant should be fine, but check with your advisor and/or committee. They might have restrictions and/or disclosure requirements for you to consider.

    It sounds like you proposed the wrong type of statistical test for the questions you're asking and the hypotheses you're testing. Frankly, but you should know these things by the proposal phase. Again, a consultant will get you lined up.
     
  4. Randell1234

    Randell1234 Moderator

    I hired a consultant at the end to help with how to present the data. The committee asked me several questions about WHY I did certain things or WHY I used the statistical methods I used. The bottom line is know your stuff and your results. Do not put it all in someone else's hands.
     
  5. Rich Douglas

    Rich Douglas Well-Known Member

    My first doctorate was done with a quantitative study. Because I used some fairly complicated stats--ANOVA followed by Tukey-T tests for individual pairs--I felt a need to compile and manipulate the data myself. I felt closer to them that way and could discuss them at will. I used a computer program to do my calculations, but I generated, and argued the meaning of, the data. Randall1234's advice is extremely important. If you get help, fine. But you own it at the end of the day. In fact, the questions he suggests are the ones you'll face. No one is going to come behind you and check your maths.

    While I'm ranting, let me add one other thing for other interested readers. In both my doctoral programs, I met a lot of people who chose to do qualitative studies because there were afraid--literally--of the statistical requirements of quantitative analysis. This is, in a word, stupid. I've done one of each. The qualitative approach is WAY harder. It wasn't even close. And this is after I had already written (successfully) a doctoral dissertation. The simple, deductive approach of quantitative analysis is much easier. That said, the approach you take should be driven by the questions you want answered, the nature of the problem you're studying. Not your preferences or your limitations.
     
  6. JeepNerd

    JeepNerd New Member

    The test I was proposing, the z test, appeared to me as the correct test, however they are telling me that it would not be valid since I am using non-probability sampling.

    My data is coming from two groups, the first is existing loans, approx 225 small business owners, where I will seek voluntary participation in my survey. (Actually doing mixed mixed method, so some interviews as well, just FYI.)

    The second group is future loan participants, the survey will become a standard follow up in the closing process.

    The population is very small, say 60 of those 'future' loans and hoping for at least 30 or more out of the 225 from the existing loan group.

    From those two groups (voluntary, mandatory) they will be separated into two more groups, those with good / poor loan performance. I am then attempting to compare the loan performance to the attribute I am capturing via the survey.

    One of the reviewers has suggested regression analysis, but honestly, I need a statistician to help me understand why the z test is not appropriate and how to fix this.
     
  7. Randell1234

    Randell1234 Moderator

    If I remember right (and I could be wrong), a z test is not "enough". I wanted to do a correlation and was told that I needed some additional measure to add more to it. I had to add the coefficient of determination. As far as what Rich said about qualitative vs quantitative, I know someone doing a qualitative and can not imagine why someone would go through that HELL! The numbers are just easier.
     
    Last edited by a moderator: Nov 28, 2015
  8. Randell1234

    Randell1234 Moderator

  9. edowave

    edowave Active Member

    You might want to try - Statistics.com - Engage an Expert
     
  10. Rich Douglas

    Rich Douglas Well-Known Member

    A z test isn't "enough" or "not enough." The question is what is being sought, and which method is being employed to seek it out.

    A z test compares two sample means to see if there is a big enough difference between them in order to infer that there is a difference between the two populations being sampled. Not enough of a difference--based on sample size, mean, and standard deviation--in the samples means you don't have enough evidence to infer a difference between the populations--that they are, in fact, the same population. Significant claims require significant evidence. This is why large inferences about populations can be made with rather small samples.

    A t test is similar, but can be used in sample sizes of less than 30. This is a non-parametric test, meaning you're not assuming the populations are normally distributed. Simply put, you need even bigger differences between the samples to infer there is a difference between the populations. It is like with the z test, which is used on samples larger than 30 and, thus, normal distribution can be assumed because, even if the population is NOT normally distributed, it behaves as if it is when sample sizes are that large, so they can be compared using a z test. A t test used on sample sizes greater than 30 gets the same result as a z test, so the simpler z is used. (Computers make all of this easy now.)

    Analysis of Variance, ANOVA, is used to compare multiple samples. It tells you whether or not there is a difference somewhere. If there is not, no need to search further. If there is, post-hoc tests (like the Tukey T) are available to ferret out the difference(s).

    Correlation compares the impact of an independent variable on two samples (dependent variable). If they move in similar directions, they are said to be correlated. If they move in opposite directions similarly, they are inversely correlated. If they move in a dissimilar fashion, they are not correlated. Correlation is not the same as causation, but correlated items are often so because they have the same root cause. You have to figure it out and argue it logically. There is no standard for correlation--how much the two samples are alike. You have to argue it. The scale is between 1 and -1, with .3 typically being accepted. If the two results are correlated to .3 or more, one (typically) can infer the two populations are correlated.

    Inferential statistics imply a positivistic perspective on how the world is to be perceived. I'm not so sure they're all that valuable outside the "hard" sciences. Social scientists strove for years to be like their older brother by applying the same methods. The constructivist movement has helped us move away from such stilted thinking as applied to humans and the social systems they construct. This has called for other methods to be employed that are just as rigorous, but not built on maths.

    Okay, I'm tired. My statistical knowledge is very rusty. Ask me about qualitative analysis, especially Grounded Theory. No, wait, I don't have an hour to explain it. Read Constructing Grounded Theory by Kathy Charmaz.
     
    Last edited by a moderator: Nov 28, 2015
  11. edowave

    edowave Active Member

    One of my classmates did a qualitative dissertation using Grounded Theory. Wow he had to do a lot of work! After seeing what he went through, I knew quantitative was the way to go.

    With all the statistical software packages out there today, a quantitative approach isn't that bad. The hard part is choosing the right statistical test and collecting the right data. After that, its almost just push a button. JeepNerd, I'm curious to know which software package you are using.

    I vaguely remember a book or a website out there called "How to choose a statistical test." It basically had a bunch of flowcharts that asked about your data. You followed the flowchart, and it told you what test was appropriate and explained why. This might help also Choosing the Correct Statistical Test in SAS, Stata and SPSS

    If you really want to know the ins and outs of the approach you are going to use (for your defense, for example) I found those little green statistics books from Sage Publishing to be a great help. From when I took my stats class, to when I actually started my dissertation was about a couple of years, so I found them to be great little reviews. Just look up "Sage Quantitative Applications in the Social Sciences" on Amazon and you'll find a Sage book on just about any approach you would want to use.
     
    Panama_PHD likes this.

Share This Page