"Descriptive analyses were conducted to evaluate patterns in demographic characteristics of the sample. The first hypothesis, which assessed whether there was a significant difference in the perceived level of victim disclosure of drink spiking experience between the United Kingdom and Iceland, was examined with chi-square. The assumption of homogeneity of variance for the chi-square was violated".
This is because the group sizes were vastly unequal, what do I do with this? What solutions do I have?
Oh, thank you!
Statistics is clearly not my strong suit.
For good advice, a little more information on the test conducted would be helpful. ... It sounds like you have a contingency table of counts ---- perhaps 2 x 2 ( UK / IS vs. Yes / No) ---- and that a chi-square test of association (or chi-square test of independence) was conducted. ... If so, the difference in sample sizes between the two countries shouldn't be a problem. Unless the "expected"* counts in the table are low.
___________________________
* "Expected" counts in a chi-square analysis on a contingency table is a technical term.
Following from Sal Mangiafico 's comment, it sounds like whatever statistical procedure you are using ends up comparing some value with the Chi-Squared distribution. A lot of tests have this aspect. What is your unit of observation? Were people sampled at multiple sites (e.g., is the design people nested within clubs?)?
p.s. I assume it is not a test of no association of a contingency test because Bjarndís Sara Breiðfjörð Anbari says the variances are different. What values do you have for the variances?
Q. How was "perceived level of victim disclosure of drink spiking experience" measured?
I appreciate the responses :) I have no idea what I'm doing!
I think I figured the chi-square test out, and as the expected value was greater than five in 80% of cells in one part... It was adjusted by looking at the likelihood ratio for significance testing.
The perceived level of disclosure was measured by requesting participants to respond to a few drink-spiking incidents in the way of saying what the likeliest outcome for the victim in each case would be, that is, 1. disclose of sexual assault to the police and seek support services, 2. seek support services only, or 3. suppress the sexual assault within oneself.
Now that I've hopefully resolved the issue with the chi-square correctly, I'm faced with a new dilemma...
In the last part of my results, I conducted mediation analyses with the process tool by Andrew Hayes to see whether perceived social support would result in greater perceived disclosure.
Because I needed to write up if any assumptions were violated, I ran linear regression separately. What assumptions are required to be met for regression to run smoothly?
For example,
The normality is violated as I have unequal group sizes but could I still include the regression analyses in my dissertation, or would I need to use another analysis? what alternatives would I have if that's the case?
Thanks in advance 🙏🏼
IMO you need far more help than you will get from a few exchanges in an online discussion forum. I suggest that you find someone knowledgeable in your department at Bournemouth University to help you. There must be someone teaching statistics in the grad programs, for example. Good luck.
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology