Dear Marian, your sample is large enought to no justify the application of a t-test Analysis (N < 30 sbj).

I sounds like you want to do an Analysis of Covariance (ANCOVA), which is basically an ANOVA which also contains control variables -- with ANVOVA because you have three categories in you independent variable.

Hi Marian, it seems to me that you have 3 variables that are the 3 hospitals, in which case you should continue your analysis by ANOVA. Or as the colleague said above, by ANCOVA.

The fact that the sample> 30 gives you assurance to assume that the sample follows a normal distribution, by the central limit theorem, so, it is not necessary to use the non - parametric methodology.

I did not quite understand what the doubt about the t test could explain better?

Best regards.

=================================================================

Olá Marian, me parece que você possui 3 variáveis que são os 3 hospitais, nesse caso, deveria prosseguir sua análise pela ANOVA.Ou como disse o colega acima, pela ANCOVA.

O fato da amostra >30 te dá segurança para assumir que a amostra segue uma distribuição normal, pelo teorema do limite central, e portanto, acaba não sendo necessário lançar mão da metodologia não - paramétrica.

Não entendi muito bem qual a duvida do teste t, poderia explicar melhor?

Melhores cumprimentos.

Hi Marian, your sample size is large and there are three hospital to compare, you can use ANOVA.

Hello,

I have 3 hospitals but if I use a combined sample and want to check hospital progress (all 3 together) on certain categorical variables --- each with 2 or 3 groups.. should I use t test only for categorical var with 2 and anova for 3.. some have even advised me to use Z test for 2 (given sample size is large)..

Taking into account your last post, I disagree with who advised you and recommend MANOVA, which is the multivariate ANOVA.

Levando em consideração sua ultima postagem, discordo de quem te aconselhou e recomendo a MANOVA, que é a ANOVA multivariada.

Basically ANOVA is conducted using F-distribution. For F-test, one need not worry about the sample size. Hence, the question of having 28 observations and 100 observations do not come into picture. Similarly, for t-tests also. Both tests are valid when the sample size is less than 30.Also,they are valid when the sample sizes are large. Z-test can not be conducted for smaller sample sizes. In my understanding, a two way ANOVA with hospitals as rows and other groups ethnicity, gender etc as columns may solve your problem.

Good luck

Thank you all for these responses.

Hi,

I agree with those who suggested MANOVA (perhaps, you could also run a regression model). Please note that, differently from what a previous comment stated, you actually need an adequate sample size to test for those models' assumptions.

By the way, I would like to add that you cannot use t-tests when you have three or more groups: doing so would inflate the probability of false positive results. The probability of not rejecting the null hypothesis when the null hypothesis is true (and the assumptions of the test are met), at a significante threshold alpha, is (1-alpha); if you have three groups to be compared, then you'll have to do six tests. The probability of not rejecting the null hypothesis in all the six tests would be (1-alpha)^6, so the probability of rejecting such hypothesis would be higher than alpha. That's why we use ANOVAs (and that's why we adjust alpha in post-hoc tests).

I am not clear whether you are actually dealing with multiple dependent variables that you want to assess as a single set, which would be the reason to MANOVA. In particular, if have data on your dependent variable at two points in time, it would be possible to deal with that by creating a change score (post-test minus pre-test) and then using ANOVA.

"I have 3 hospitals but if I use a combined sample and want to check hospital progress (all 3 together) on certain categorical variables --- each with 2 or 3 groups.. should I use t test only for categorical var with 2 and anova for 3"

Marian Abraham, when you say you want to check hospital progress "on certain categorical variables", it sounds as if those categorical variables are dependent (or outcome) variables that have been measured at least twice. But then you talk about t-tests and ANOVA, which suggests that your outcome variables are quantitative. I can't speak for others, but I'm confused. Please tell us what the dependent (outcome) variables are, how they are measured, and how often they are measured. Also give more detail about the "certain categorical variables". What are they? Are they explanatory variables or outcome variables. Thanks for clarifying.

Finally, I think it would be wise to include hospital as a categorical explanatory variable in your model rather than pooling the data from all 3 hospitals. Pooling the data can lead to misleading results--look up Simpson's paradox, for example.

HTH.