### BIOSTATISTICS

Transcript: Ho: Median <= zero HA: Median > zero C: when C> C.V. then do not reject null CV. of C: B.27 or page 786 in ZAR H0: A less than or equal to B HA: A > B Yes, the data seems to be normal. We have no reason to think otherwise. BIO STATISTICS Statistic H is determined by ranking all observations But, when N is too small use chisquared aprox test. H0: things are not dependent HA: things are dependent. aka one changes the other There are more than two categories. So, run a goodness of fit test. No, the data can not be paired. Rules for flow chart: First you must pick if you are testing central tendency or distribution. Pick the appropriate bracket. After that, there are questions on every slide. If the answer is yes to the question: then go LEFT. if the answer is no: go RIGHT. Used when several treatments are compared back to a single control example: drug A B and C compared to a control Yes, the data is nominal. So, run a Chi Squared Category Table. Tests for independence. Contingency Table Well, the data isnt normal. But it can still be paired. So, run a Wilcoxon Paired Sample T-Test Transform data: - Log >exponential -Square root > butted against zero Arcsine > percentage data ties: tied observations get mean ranks, they change possible outcomes Falls between Tukey and Dunnets as far as riskyness and Type 2 error Yes, one variable is in fact affecting another. Great, run a regression!! No, it is not nominal. Kolmogrov- Smirnov Test mean comparing a few pairs -two pairs: pairwise rate of 5% -so if you do too many - you're increasing your error No, the variances are unfortunately not equal. So, run a Welches Approximate T-Test. If there is a significant goodness of fit test you can do further comparisons (Sub Dividing) which indicated which category contributed to the significance. -This is exploratory Randomized Complete Block Design Yes, the variances are equal. So, run a Two Sample T-Test. p-value is significant!!! YAYAYAY are multi-comparisons planned (go left) or not (go right) Shoot. Is the data nominal? 2 SAMPLES Example where we did male vs. female and what was important in a relationship. No, I don't believe it is normally distributed. So, run the non parametric version of thr 1 way ANOVA, The Kruskal-Wallis Test or transform the data. report rsqu: coefficient of determination H0: b= 0 HA: b not equal 0 Awesome! There are 3 samples. How many factors are there? Yes, that data could be paired. But is it normal?? Student Newman Keuls No, the data is is not normal. So, run a Mann-Whitney Test. Yes it is normally distributed & there is only 1 factor. So run a 1-way ANOVA. multi-comparisons Yes, it is nominal. 1 SAMPLE H0: >= HA: < One or two tailed?? sample sizes testing distributions No parametric version test of continuous probability stats to compare a sample to a reference probability Also- plot twist: there are grouped KS as well! like the example we did in class that was Dry cat food to Wet cat food because there were 5 categories PS: if it is non-parametric then run a Spearman rank corr coeff. If significant p value: probs should do multi-comparisons There is more than 1 factor. So, a two-way ANOVA would be most appropriate. ps: there is not a non parametric version Is it nominal? Next question: could the data be paired? FISHERS LSD (least significant test) So run a one sample t-test if C > C.V. then reject null t.005,1,d.f.= chart C.V.: table B.3 page 678 in ZAR Mu: excel S (standard dev)= excel sxbar= s/squ of n t stat= graphpad 95 C.I.= graphpad d.f.= n-1 pvalue= graphad Tukey Test Rules: 1.) if n's are too small then calculated chi square os biased 2.) no expected freq should be < 1.0 3.) no more than 20% of observed should be less than 5.0 No, it shouldn't be normal. Planned= those you believe in advanced should show significant differences Is it normally distributed?? only used to analyze two or more categorical variables 1 If p=value is significant: probs should do multi- comparisions test. This is a sign ranked tests on the differences (d). H0: Median = HA: median does not equal ___ median= p= t0.05,1,n -> C.V.: B.12 or page 758 in ZAR t-: report the lesser of the two t-, t+ ^^^ 1. sort data from least to greatest 2. assign rank values based on median 3.if ties: average between the two HOW MANY SAMPLES ARE THERE? Dunnetts Test Yes, it can be paired and it is normal. So run a Paired Sample T-Test. PS: if you can pair, you should because it basically eliminates variability/"background noise" & increases the accuracy of the estimates YOU HAVE TO BE CONFIDENT YOUR DIFFERENCES ARE NORMALLY distributed -> is the n large enough? H0nutrients: MU 1 = _=_ =_=_ HA: at leave one is different H0: water H0 interaction ^^ three sets of hypotheses' YES, it is normal!! has a experiment wise error rate of 5%- conservative JB says toooo conservative.. like me lol but easy to do on JMP -this test just compares all means to each other How many categories are there? If main effects are significant that run