2

I would like to know which technique is appropriate for analyzing data collected via self-administered survey. Some background information follows:

Topic: Breastfeeding support and the workplace

Setting: University campus

Sample size: 30-40 working women who were on maternity leave from 2011-2012 (first 6 months)

Hypothesis: Breastfeeding support influences breastfeeding initiation and duration.

Example questions: socio and demographic items in order to describe women, Likert scale questions addressing: university policies and work culture, immediate manager support, immediate colleague support and work flow. There will be some close questions related to the intention to breastfeed.

This builds on my previous question about the sampling approach and what to treat as the unit of analysis.

I really appreciate all your help and consideration.

Best,

Community
  • 1
Wendy Alfaro
  • 129
  • Survey sampling inference can be applied if the sample is randomly selected from a population. If not there may be no way to know what biases are introduced through the sampling mechanism. – Michael R. Chernick Jun 12 '12 at 02:36
  • Yes, the sample will be randomly selected from a population. – Wendy Alfaro Jun 12 '12 at 02:50
  • I've edited your question to add a link to your previous question as further context. – Peter Ellis Jun 12 '12 at 04:07
  • Sample size will be quite a challenge for you. 30-40 is a small group given the effects might be quite subtle. To get insight from this, I would suggest you consider qualitative approaches at least as an addition to any statistical testing. – Peter Ellis Jun 12 '12 at 05:47

2 Answers2

1

My guess is simple regression methods will go a long way towards helping you understand your data. Although this is not usually considered an advanced statistical topic, if you are not already sufficiently familiar with these techniques you will have a steep hill to climb. In such a case I would suggest working with a good statistical consultant; the university probably has some. This is especially true since @MichaelChernick's point is on target, and that it sounds like you want to make a causal inference, but will have observational data, which is, in fact, a very advanced statistical topic.

gung - Reinstate Monica
  • 145,122
  • Thank you so much Gung. Please let me know if you can recommend me some articles to read. I am ready to learn as much as I can. – Wendy Alfaro Jun 12 '12 at 03:33
  • I'm going to assume you're asking for something to help you make causal inferences from observational data. You might want to try Rosenbaum's Observational Studies, an easier start might be Mostly Harmless Econometrics. – gung - Reinstate Monica Jun 12 '12 at 04:18
  • According to Wendy it is a random sample and not observational data. The term self-administered really just means that she adminstered the survey and created it. It looks like correlation between the responses and when the breast feeding was initiated and how long it lasted. This could possibly be a regression analysis as Gung suggested. – Michael R. Chernick Jun 12 '12 at 04:49
  • @MichaelChernick, maybe I'm confused. I was thinking of a random sample as licencing inferences from the sample to the general population that she really wants to be talking about, but that by 'observational data', I was assuming that she wasn't randomly assigning new mothers to different levels of breastfeeding support that she had independently manipulated, and thus she cannot directly apply a causal interpretation to her findings. I think of these as distinct issues. Am I missing something here? – gung - Reinstate Monica Jun 12 '12 at 05:04
  • @gung It could be me who is misreading the question. I have not seen hte previous post that she refers to. As far as causal inference is concerned, it is always difficult to attribute cause and effect to observed relationships regardless as to how the sample is taken. Observational studies such as case-control studies are designed to eliminate potential bias through matching pairs. – Michael R. Chernick Jun 12 '12 at 10:24
  • Thank you so much to all of you for your feedback. Please let me know how can I clarify any question that you may have. – Wendy Alfaro Jun 12 '12 at 12:15
  • I will be selecting the sample randomly. After that I will send the selected mothers a survey via mail, email or a link to the html version. I also will be sending a modified version of the questionnaire to the immediate managers and coworkers of these women. Also, please let me know if there is an online tool that I can use writing the survey and then sending the link to the participants. – Wendy Alfaro Jun 12 '12 at 12:20
0

You can compute average scores and draw inference about the population means based on the formula for the variance of the estimate which is f σ$^2$/n where f=1-n/N is the finite population correction N is the population size, n is the sample size and σ$^2$ is the population variance. Sometimes questions are designed to represent certain domains and the response can be combined to get domain scores. If that is the case inferences to the population can also be made based on those combined scores. Cochran's classic book Sampling Techniques provides the most lucid description of the theory, See this link: http://www.amazon.com/Sampling-Techniques-Edition-William-Cochran/dp/047116240X/ref=sr_1_1?s=books&ie=UTF8&qid=1339472916&sr=1-1

Michael R. Chernick
  • 42,857