I am trying to analyze restaurant ratings and comparing it by the different cuisines. The problem is my data set has uneven sample sizes (6 Greek rest., 100 Indian, 50 Lebanese, etc.). How can I take this into account when looking at the reviews?
2
-
$\begingroup$ how do you want to compare them? What is the question you are looking to answer? $\endgroup$– ReneBtAug 13, 2018 at 8:58
-
$\begingroup$ I am comparing the different ratings for example. On average Greek restaurants get a rating of 4 while Indian get a rating of 3.5. However the data set only has only 6 Greek restaurants and 100 Indian ones. How can I compare all these different cuisines with different sample sizes. $\endgroup$– AbdallaAug 13, 2018 at 9:27
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
2
Depending on what statistical test you are trying to use, in many cases a difference in sample size or even a small sample size will not actually stop you from carrying out the statistical test. However, you are unlikely to get a 'statistically significant' result, because your sample size is too small to detect a difference even if one exists.
A pragmatic way to get round this may be to combine some of your categories of restaurants into larger groups. Your choice of groups is likely to be subjective, but it would be one way to even out the sample size of your groups. Or only compare groups that you have a reasonable sample size for.
It's difficult to say more without knowing what statistical analysis you're trying to carry out. Your answer to ReneBt's comment suggests that you may be dealing with scores that can only take fixed values along a rating scale - e.g. 3, 3.5, 4. Have you taken this into account in your choice of analysis?
-
$\begingroup$ Izy, thank you for your response. What I am trying to do is determine what cuisine is rated the highest in a specific city. So I am averaging all the scores of a specific cuisine together. I have eliminated any cuisine with less than 5 restaurants. However I still end up in a situation where some of the highest rated ones only have 7 or 8, and some average rated cuisines have over 100. I am just trying to figure it what is a fair way to do this comparison. I hope this clarifies what I am trying to do. $\endgroup$– AbdallaAug 14, 2018 at 7:22
-
$\begingroup$ So are you trying to test if there is a significant difference between the mean values of different types of cuisine? In that case then an ANOVA would be a simple way to test this. However, if your data can only take fixed values along a rating scale, then you may want to consider an ordinal regression model instead. However, a sample size of 7 is very small - possibly too small to find a 'significant difference' even if the averages look different to you. Have you considered that 'rarity' may boost the scores? Perhaps try putting all rare cuisines into one group instead of eliminating them? $\endgroup$– IzyAug 15, 2018 at 9:24