For my bachelor thesis, I'm seeking to prove that the degree of adoption of value based pricing strategy has a positive effect on the innovation performance of SMEs. At first, I thought to use a regression analysis. However, given the low response rates, I only have 12 respondents. Therefore, what i'm asking is, can i still do a regression analysis in order to produce some results? (Even if not statistically significant) Or do you have any suggestion on any alternative techniques?
Hi
1. How many random variables in your research. How many dependent and independent random variables.
2. What is the relation between the random random variables, i.e. related or not.
3. Regression analysis required a linear relationship between the independent random variables and the dependent random variable.
4.If the sample size of your random variables >30 you can use most of statistical tests.
5. If the sample size of your random variables
Actually, none of the statistical tools seems good. But try to make both variable dichotomous and try Fisher exact test. Regression may not apply as your sample size is too low for prediction.
Hello, atlhough it isnt an great statistical approach helps to test the hyphotesis you can use a T-test. Just check if with 12 would be acceptable.
Hi
1. How many random variables in your research. How many dependent and independent random variables.
2. What is the relation between the random random variables, i.e. related or not.
3. Regression analysis required a linear relationship between the independent random variables and the dependent random variable.
4.If the sample size of your random variables >30 you can use most of statistical tests.
5. If the sample size of your random variables
What about dichotimizing the DV and trying logistic regression? See the attached reference. Best, D. Booth
Since your number of responses is small, use logistic regression and make your variables dichotomous.
My experience in my research on diseases and patients where sample size is less than 30 or 30 due to difficulties involved in sample collection is that the application of T-test is most appropriate
See Amir D. Aczel "Complete Business Statistics".
The answer: t-student (examples in above mentioned book)
Best regards
Hubert Anysz
Use non parametric tests.,Mann Whitney u to compare means for paired data use Wilcox test.for dichotomous variables use fisher test .
Hello, your sample is very small. Parametric tests and Regression Analysis are not acceptable. I have two suggestions:
1- To use spearman correlation to check the correlation between the variables.
2- To use Chi-Square test to verify the association between the variables.
And one more thing: Everytime that we say "influence" we are talking about regressions. And when we are saying "what's the relations" we are talking about correlations.
Good lucky!
Well, Your response rate is very low.I suggest you check your questionnaire for reliability and content validity.If after rewording the questionnaire, your response rate is still low, then, i suggest that you use t- test.
In addition to the use of t-test, I suggest you adopt more innovative methods to increase your sample, for example, your social network, referral, etc.
Hi Pippo
In order to calculate and interpret the statistical tests, the following documents need to be utilised.
Regards,
Zuhair
The answer depends upon the number of observations you have. Do you only have 12 data points? Or do you have multiple observations for each person? Suppose that each person has K observations (repeated measures). And let V be the number of independent variables. Then, for regression to work, you will need 12*K > V+1. (The "1" in this formula allows for the intercept in the regression model.) In any case, when 12*K is approximately equal to V+1, then the R-squared will be very close to one. This does NOT mean that you have a good model. It means that you have very limited information.
I am going to disagree with most of the comments above. If the dependent variable is continuous, leave it "as is". Do NOT turn it into a dichotomous variable. This just throws away information. You are much better off staying with regression -- unless there is some compelling subjective matter reason to have dichotomous variable.
Also, consider the so-called "jackknife resampling" approach to getting standard errors (that is, doing your stat tests). (There's a Wikipedia article on this topic.) Basically, you rerun the regression 12*K times, each time throwing away ONE observation. (That is, each time you have only 12*K - 1 data points for the regression.) Each time you will get different regression coefficients.
Save all these in a spreadsheet. (Your spreadsheet will have 12*K rows, one corresponding to each regression run. There will be V +1 columns, one corresponding to each variable in your regression model.) Then, compute the mean and standard deviation for each regression coefficient in turn.
Take the mean MU of each column as your estimate of a regression coefficient. Take the standard deviation SD of each column as the standard error of each coefficient. Then, compute Z = MU/SD as a rough index of statistical significance. Coefficients for which the absolute value of Z is greater than 2 are probably important.
Finally, the jackknife resampling procedure can be used on ANY statistical model. For example, if you decide to go ahead and do a logit regression model, you could also compute standard errors using the jackknife approach.
Your problem is an interesting and difficult one in statistical inference. Good luck with your research!!
Gabriele, I’m not agree with most of the answers of the colleagues in this forum. If you didn’t planed your research with an adecuate number of samples considering the variables, the characteristics of the population, variability, significance level, power and size effect, nothing or almost nothing you can perform with the results thinking in apply inferential statistics. This is the major problem with non-planned research, you must think about the analysis previously you perform the research, if not, there is not an appropriate statistical solution. Statistic have many critics and the main reason is due to responses like these (you can see many different approaches in this forum), even some that reveals no statistical knowledge (sorry).
But what you can do? I recommend to perform an exploratory analysis, concentrate in descriptive statistics including graphical analysis and don’t try to generalize your results with inferential statistics (even non-parametric). That is better and more honest with your data, your research and the rest of the scientific community.
Good luck.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics