Choosing a Statistical Test for Your IB Biology IA

[@danielm9463]

One small caveat: in broader mathematics, "number of bacterial colonies" would be treated as a *discrete variable*, which means the variable is numeric but it's restricted to certain values (and between those allowable values are gaps that the variable can't take on). But if you're plugging that variable into a regression or t-test/ANOVA model, then you're treating it as continuous. To quote minitab, which has great articles on statistics: "If you have a discrete variable and you want to include it in a Regression or ANOVA model, you can decide whether to treat it as a continuous predictor (covariate) or categorical predictor (factor). If the discrete variable has many levels, then it may be best to treat it as a continuous variable. Treating a predictor as a continuous variable implies that a simple linear or polynomial function can adequately describe the relationship between the response and the predictor. When you treat a predictor as a categorical variable, a distinct response value is fit to each level of the variable without regard to the order of the predictor levels. Use this information, in addition to the purpose of your analysis to decide what is best for your situation." support.minitab.com/en-us/minitab-express/1/help-and-how-to/modeling-statistics/regression/supporting-topics/basics/what-are-categorical-discrete-and-continuous-variables/

[@madipsychomadi3144]

thank you very much for your generous response.

[@felixokatch4587]

Good

[@danielm9463]

​@Raggul Kanakasabapathy, it sounds like it could go either way. I'm not an expert by any means, but I think you would decide whether to treat the independent variable as categorical vs continuous by asking questions like "do I expect a linear relationship between the input and output, such that more of the independent variable consistently leads to more/less of the dependent variable?" I think you can also ask questions like "for the purposes of my study, do I care about the *relationship* that the independent variable has to the dependent variable, or do I care about finding e.g. a single dose that is therapeutically significant?" Do you have reasons to expect that salt has a linear relationship (positive or negative) with plant growth? Are you interested in the specific nature of that linear relationship (e.g., is your goal is to say "for every 1 g of salt, the leaves grow X cm longer")? If so, then it might make more sense to treat mass as continuous. Alternatively, are you looking for whether a particular mass of salt exists which provides a statistically significant growth effect? Are there reasons why too little salt could inhibit growth, and too much salt could inhibit growth, but maybe there's a salt mass in the middle that could balance multiple biological factors and yield the most growth? If so, then it might make more sense to treat mass as categorical. The fact that you described 0 g of salt as the "control group" already suggests to me that you might be looking to compare salt vs. no salt, and the reason you have different amounts of salt is to help answer the question "do plants grow more effectively with salt than without salt?" In other words, having a control group suggests that you're not looking for the specific nature of a hypothesized linear relationship between amount of salt and amount of growth. I've never taught IB Bio, but I did teach IB Physics. If you're writing an IB Biology IA, I would think it's worthwhile to explain how you're analyzing your data and justify why you chose to treat the independent variable as categorical vs. continuous. If you choose to treat mass as categorical, then you've got a categorical independent variable and a continuous dependent variable--what test would you use in that circumstance? If you treat mass as numerical, then both the IV and DV are continuous--what test would you use in that circumstance? (By the way, you can still do a regression with a categorical independent variable, but that's not really done in IB Biology, and it sounds like this might not make sense in your case. Here's what it would entail. For each dose, you'd essentially have a different dummy variable that takes on a value of 0 or 1. So the variable m_1g would be 0 or 1, and the coefficient would tell you how much more/less growth the plant had compared to 0 g when 1 g of salt was added. The variable m_2g would similarly take on values of either 0 or 1, etc.) Another quick thing--if you go the route of performing multiple t tests (to compare each salt mass to the control group), then you want to watch out for family-wise error. You can do a quick Google search to see what this is. This is usually dealt with by lowering the p value that you need in order to reject the null hypothesis. For example, if you do 5 t-tests, and you'd normally use p < 0.05 as your significance level, then you might instead calculate 0.05/5 and use p < 0.01 as the significance level for each test, since performing multiple tests increases the likelihood of a false positive. There are different ways to account for this family-wise error, and the appropriate approach depends on the class you're in and how rigorous the statistics element is.

[@danielm9463]

@Raggul Kanakasabapathy Since it's an IA, I can't quite give you a direct answer--that will be up to you. All I'll say is that it sounds like you're saying you have a continuous IV, a continuous DV, and you seek the mathematical relationship. if you go the route of regression, I have a few thoughts: (a) I think regression and mathematical relationships are less common in IB Bio IAs. Because of that, I recommend asking your teacher if you're on the right track and if you have a suitable topic, question, and statistical analysis approach. (b) If you go the route of regression, I think you will want to be careful about how you phrase your research question. If your understanding is that regression is appropriate when seeking a mathematical relationship between a continuous IV and a continuous DV, then you want to phrase your research question as seeking that *mathematical* relationship between the variables. (c) In all of it, the ultimate goal is to use the analysis as a mechanism to talk about in-depth Biology analysis. So if you find a negative linear relationship, then you want to use the Biology you've learned to try and explain, in depth, *why* that relationship exists. (d) Since I've never taught IB Bio, definitely have a conversation with your teacher about all of this. The advice I'm giving you is good for IB Physics, but I'm not sure how much of it translates over to Biology.

[@danielm9463]

@Raggul Kanakasabapathy Hi Raggul, I was curious how your IA turned out! Were you happy with the final product?

[@tekmepikcha6830]

This is perhaps the most simpliest and yet full tutorial I've heard on statistical tests LOL.....Well put together Daniel. Thanks

[@danielm9463]

Thanks for the kind message! I'm glad it was helpful

[@mariabelenvargassanmartin2314]

FACTS

[@RG-sl6ix]

Finally, the sky is clear....

[@robingriffin42]

9:57 minute video vs 6 weeks of lectures...just wow!

[@14MCDLXXXVIII88]

This is by far the best material about SPSS I've ever seen in my whole life. Throw 100$ SPSS books into trash can. This video is very beautiful in a minimalist way. Greetings from a procrastinator doctor, who is trying to hurry analysis at 07:00 AM for the finish date of a paper :-)

[@Tiatabs13]

I have all this data I’ve collected for my masters thesis, I’ve been looking for a test to use for two days. This video, hands down, has just saved my life. THANK YOU!!!

[@jaleelhull941]

how did it go

[@view5558]

After struggling for years trying to figure out about the necessity of so many statistical tests, finally I have an overview of statistical tools an how to choose one from.

[@rodsalomon6524]

Im from the Linguistics field in this is still helpful for my thesis, thank you!

[@mmmm7656]

Thanks to Allah.. I have found someone who is teaching the basics..I needed it badly

[@asmasultana2732]

Hello. I need some help from you

[@mmmm7656]

@@asmasultana2732 yes

[@danielm9463]

I'm glad it helped! I often need the basics too. This is a good starting place, but this video really does make some simplifications, and as you study more, you'll see that things aren't always so clear cut. I'm glad it helped!

[@gutsandglory7507]

Man, this guy nailed it. I fumbled in my undergrad research. I wish this was available to me then. Thank you, Sir.

[@Lydia-yo4lo]

Thank you so much for this video! Simple, concise, well organised -- it's rare to see such a well-made tutorial to a somewhat confusing topic such as this, amazing :)

[@asmaaadventures1791]

Wow! I cannot express how much I am grateful to you for making this video. I spent one week figuring out which suitable statistical test was for my case. THANK YOU

[@view5558]

After struggling for years trying to figure out about the necessity of Sr many statistical tests, finally I have an overview of statistical tools an how to choose one from.

[@danielelieh-ali-komi6544]

This video includes more data and has better educational content than what I learned in my MSc. Thank you Daniel.

[@1990poulami]

I watched so many videos on this topic. But this is the best explanation so far. Thank you so much

[@kteforeal]

I'm slow when it comes to math (like "if johnny has 5 apples" type slow) and this breakdown finally made things clear

[@danielm9463]

Don't hesitate to post any questions you have about your work or what you're studying!

[@v9054]

I fell in love with the way you explained both the qualitative and quantitative techniques.

[@danielm9463]

Thanks Venu!

[@jodieharth]

Thank you! Was recommended for Business Data Analytics. 4:08 is particularly useful if you're wondering what sort of test to use (or, what it's called!) Chi-Squared / t-Test / Correlation.

Haters will say this clip is way too simple, I will respond yes that’s why it’s almost perfect.

[@darrenzerone765]

Statistics are so simple and easy to understand after your wonderful explanation. Thank you for this amazing video.

[@chaqsakhatanachaqsakhatana2325]

Stat is not easy subject 😭😒

[@saggrawal]

An amazingly simple description of statistical tests. Thank you so much!

[@happyshappys]

Simple, easy and best and proper way to have quick notes

[@donharris8846]

For a broad perspective, this is a remarkable video. Nice work! This helps a lot

[@danielm9463]

Thanks so much, Don!

[@michalinaprycka3692]

This video has helped me more than my biology teacher, love you man

[@danielm9463]

Glad it helped! I sympathize with IB Bio teachers! They're expected to teach statistics, but it's not part of the curriculum!

[@Sarmilagiri12]

The most simpliest and useful tutorial i ever heard and seen. Thank you soo much for giving us such a wonderful lessons. 😊😊😊

[@yusronnaim]

half semester of my statistics class didnt give such comprehension, thank you for this well put video.

[@m-tron5966]

The Simplest and Most Comprehensive Demonstration of Tests...Thank You

[@maria_yourKARMA]

I like how the simplicity of your content, keep it up.

[@kaliswarijawacar1501]

I am not clear about which technique is used for the which study before observing this video. Now I clear about it . Thank you sir

[@NileSon123]

Best video I watched in statistics especially in experimental designs Thank you very much for that simple and valuable information

[@novaxgreed2828]

This video basically summarized two semesters of my statistics subject ngl.

[@tameka4100]

your english is impeccable. i can clearly hear the proper pronunciation and enunciation. Concept is well explained.

[@danielm9463]

It's funny--I think people think I'm smarter than I am because I enunciate well :)

[@ottobena]

Simply the best and clearest explanation video.

[@marcomurru3295]

Is Chi Square not used when you want to find a relationship between two or more variables? I'm a bit confused now.

[@danielm9463]

It's a really good question. In this video, I'm using "relationship" to have a narrow meaning. It refers to relationship in the mathematical sense, which is the nature of the equation that relates the two variables. It's similar to how "correlation" has narrow meaning and only refers to *linear* correlation. The Chi squared test isn't about determining the specific nature of the mathematical equation relating the variables. If we stick to dichotomous variables (for simplicity), then the Chi squared test of independence is generally used when we have 1 sample, and we measure 2 categorical variables, and we want to know if they are independent or associated. (Just as a side note, "associated" is different from "correlated." Correlation is one specific type of association wherein the relationship is linear.) The Chi squared test of homogeneity is very similar. Generally, we can think of those problems as having 2 samples, where we measure the same categorical variable within each sample and we want to know if the two groups are the same or different. So neither Chi squared test assesses the specific mathematical nature of the relationship, which is why I'm not classifying it under the narrower mathematical definition of "relationship" in the video.

[@datapeek]

I watched many videos on Hypothesis tests and always left with some sort of confusion. But you made it really simple. 🔥

[@s1lverhawk]

From the bottom of my heart THANK YOU for this

[@solomiiak1779]

Thank you sooo much for this video! It saved my life!I spent so many days trying to understand which test I should use for my master's thesis. You explained VERY well and simple! Thank you again.

[@_kaira4130]

The way to generalize test in a table is amazing!! Thanks! It really really helps a lot!

[@danielm9463]

Glad it helped! Just don't miss the correction I mentioned in the description

[@tessaowens]

Correlation can be used on categorical data too! Not just continuous data.

[@jincymathewm362]

Yes, spearmans rank correlation

[@mrkhan.881]

Yes. But the dependent variable should always be continuous variable. And if the independent variable is Nominal then use Phi coefficient test and if ordinal then use Spearman.

[@danielm9463]

@@mrkhan.881 That's true for linear regression. Other regressions (e.g., logistic regression) do permit the dependent variable to be categorical.

[@danielm9463]

@Tessa Owens, it's a good point. Really, a lot of the dichotomies in this intro video disappear in higher-level stats. You can reproduce the same p-values in Anova as a regression model, and you can set up dummy/indicator variables to encode categorical variables. The main audience for this video is high schoolers who haven't learned any stats but are expected to use it in their IB Biology IA.

[@danielm9463]

@@jincymathewm362 It's a good point. I've intentionally avoided ordinal data in this video because it really is a separate category from a stats perspective and usually requires nonparametric methods. Fortunately IB Bio high schoolers aren't expected to use it!

[@naziya9967]

Iam from india , teaching research to post graduate students of education department.feel very helpful to understand overall category /families of statistics test.very excellent content of video 👍👍👍👍

[@flyingfox707b]

Very succinct and well put together. Thank you, Daniel!

[@ikmalhakim1101]

How do you proceed from one-way ANOVA? How do you figure out how many and which one is statistically significant?

[@danielm9463]

One approach (though I'm not sure the extent to which this is done) is to proceed with individual t-tests that run through all possible pairwise comparisons. However, doing this introduces family-wise error. For example, if you do 10 follow-up pairwise t-tests using a significance level of 5%, then the cumulative chance of a false positive in any of the 10 tests is 1-(1-α)^10 = 40%. If the true Type 1 error rate is 40%, then it would be inappropriate to claim that our significance level is α = 5%. So the significance level must be adjusted to compensate for the family-wise error. One approach is the Bonferroni correction, which simply entails dividing the significance level α by the number of pairwise tests that are planned (or conducted post-hoc). If a cumulative Type 1 error rate of 5% is desired, then a significance level of 0.05/10 = 0.005 would be used. The more precise formula gives us the exact cumulative Type 1 error rate for 10 tests each using α = 0.5%. It is: 1-(1-.005)^10 = 4.9%.

[@mainecervillon5437]

Thanks for making it super simple, clear, and easy to understand!

[@razwanulislamasif1190]

best ever discussion in truly simple way. love from Bangladesh🇧🇩🇧🇩🇧🇩

[@marjiehoss1498]

I wish you taught me Econometrics. Thank you for this.

[@kerrielashley]

Thank you for this video. You did a great job of explaining the various statistical tests and their applicability.

[@shadowmonarch809]

How about if our study is fnding the relationship of modular learning but it is categorical and 1 continuous which is the academic performance? Will we use correlation or T-test?

[@danielm9463]

Check out the summary at 4:08 of the video. You have a categorical independent variable and a continuous dependent variable. You're comparing the categorical groups (the modular learning) on the basis of the continuous outcome (academic performance). Which do you think is more appropriate--t-test or correlation? When you get into higher-level stats, you'll learn that actually a t-test can be replicated in a regression model. But for now it's helpful to think about whether you're (a) testing if two groups are different on the basis of their continuous outcome variable (that's a t-test, or comparing groups) or (b) evaluating the strength of correlation / the quantitative effect of your predictor on the outcome (that's correlation/regression, or looking for a relationship).

[@WrenStanchen]

This helped me a lot, and I'm taking a graduate level class. Thank you!

[@arushikumar3044]

This video is so well made! Thank you for your effort. It's really helping us out.

[@danielm9463]

Very glad to hear that! Check out the caveats I pinned at the top of the comments. Feel free to ask questions!

[@justiceoforiatta6826]

Thanks for enlighten me on how to choose these statistical tests. I will be grateful to learn more.

[@scarsmakestars]

I'm in IB Sports Exercise and Health Science but this video helped a lot for choosing the test for my IA. Thanks!

[@sowmyaramaswamy6686]

So helpful. Thank you! The basics make it easy to understand everything

[@kenmakozume4781]

I'm homeschooled and we were tasked with an experimental quantitative research. It is my first research and I'm all by myself. Thank you so much for your simple explanation! You saved me big time. My teacher dosent teach and her modules are hard to understand. Now, I have an idea what statistical analysis I should use for my experiment. Many many thanks again🙏

[@danielm9463]

Feel free to post a comment if you have any specific questions about your research!

[@kenmakozume4781]

@@danielm9463 My research is about aquaponics, how fish waste affects the growth of a plant, specifically Gotu Kola. My experiment will use two containers with different number of fishes to manipulate the amount of fish waste and both have one sample of Gotu Kola. Should I use correlation or t test? Thank you for you Ur reply :)

[@danielm9463]

​@@kenmakozume4781 This sounds really interesting. There might be a couple ways you could take it. Do you have a way to measure the mass/volume of fish waste, or is the plan to count the number of fish? I'm also curious how you're measuring plant growth--is this done on e.g. a daily basis or at the end of a fixed study period? One rule of thumb is that you need ~10 points minimum for correlation or simple linear regression, although you can often get away with fewer data points. ("Simple" linear regression just means there's a single dependent variable and a single independent variable.) So t-test might tend to be more natural for that reason, unless you're planning to collect repeated measures for each tank. If you do a t-test, you're treating your independent variable as categorical, almost as though you simply labeled one "high waste tank" and the other "low waste tank."

[@kenmakozume4781]

@@danielm9463 thank you for the kind compliment. I measure the fish waste amount by weighing scale(grams) and by counting the fish but the scale is more specific. I will be observing the samples for four weeks. Particularly, measuring the diameter and height of the largest and smallest Gotu Kola. Is that a t test??

[@danielm9463]

@@kenmakozume4781 it sounds like you might have two different independent variables, and possibly two different dependent variables. You'll either have to pick one of each, or simply have a few different research questions, each of which you answer separately through a different statistical analysis (e.g., #1 does mass of fish waste impact diameter of..., #2 does mass of fish waste impact height of..., etc.). Are you measuring the samples periodically (e.g., at 0 weeks, 0.5 weeks, 1.0 weeks, 1.5 weeks, ... 4 weeks), or are you measuring just at the beginning and the end (i.e., at 0 weeks and at 4 weeks)?

[@leoli9280]

is there a disadvantage to using the non-parametric counterpart to one of the tests? does IB accept and treat the ANOVA test and the Kruskal wallis test the same?

[@mrkhan.881]

Non Parametric tests are weak tests, thats why always look for normality of the data and use Parametric one.

[@danielm9463]

@@mrkhan.881 This isn't true in general. In some circumstances, the non-parametric test has a high statistical power. It really depends on the context. See, for example, stats.stackexchange.com/a/204036

[@danielm9463]

Parametric tests require that certain assumptions be satisfied. A "parametric" test is one for which we know the underlying distribution of the entire population. "Non-parametric" tests do not make any assumptions about the population distribution. So, before we can use a parametric test, we have to show that our *sample* satisfies certain conditions (usually involving normality of the data) that allow us to say e.g. "the population data is normally distributed." For beginning statistics, that's the decision criterion: if your data satisfies the conditions for a parametric test, then use the parametric test. If your data doesn't satisfy the conditions, use the non-parametric test. Many intro statistics tests will say that the non-parametric test is more flexible but has lower statistical power. This means that you can use the non-parametric version in a wider range of scenarios, but it's not as good at finding real differences when they exist. (The idea here is that non-parametric tests are more likely to make the error of saying that your results are *not* statistically significant, when a real difference does exist.) This is a fine rule-of-thumb to use when starting out, but it's not advice that advanced statisticians use, because it's not true in general. In some scenarios, the non-parametric test has a *higher* statistical power than the parametric test. It just depends on the circumstance.

[@battlemud577]

Omg... I am a year 5 dental student and this vdo finally opened my mind...

[@ainazarmstrong5615]

This video perfectly cleared everything up, thank you so much.

[@sebastianheuchler436]

Very brief and to the point. Great for a reasonable first overview of these topics. Thanks

[@anthonyoni4751]

I don't know how you did it but you have helped me to know this better.

[@arpitachopra27]

I am.a little confused. I am looking at "To what extent does Altering Cooking Methods (Boiling, Heating, Microwave) change the iron concentration (µg/dL) of Spinacia oleracea (50 grams)?". so how does different cooking methods impact the concerntration of iron concentration. I will do a spectrometer test, and find the iron concerntration of spinach using known concerntration of iron and measuring their absodance. i will createa a culliburation curve, and then use that to find relationship between the concentration of iron and the type of cooking method. what kind of test would i use to find the relationship?

[@danielm9463]

This is a great question. Reply back to this comment, and identify (a) what your independent variable is, (b) whether your independent variable is categorical or continuous, (c) what your dependent variable is, and (d) whether your dependent variable is categorical or continuous. Once you've done that, we'll take the next question together which will help you pick a statistical test to evaluate whether different cooking methods lead to different mean concentrations.

[@rohinikayal4440]

Hi I am really confused about what test to use for a research I am doing at highschool, could you help me out if you are aware which one could help me? I am testing different soil moisture percentages and it's affects on the ratio of leaves to flowers of a specific plant species. I have around 34 samples for this experiment where the IV is Soil moisture % and the DV is the RATIO (leaves by flowers). Could you suggest a statistical test for such an experiment?

[@danielm9463]

This sounds like a really cool project! I'd be interested in what your data looks like. Okay, a couple questions for you: (1) which variable is your independent variable (soil moisture % or leaf-flower ratio), and is it numeric or categorical? (2) which variable is your dependent variable (soil moisture % or leaf-flower ratio), and is it numeric or categorical? (3) which scenario is this? (check the image at 4:08) Reply back with what you're thinking (it's okay if you're not confident or wrong--give it a try and then we can figure it out together), and then I'll follow up again with some more guidance!

[@ely.romera]

Hi! I really have a question that I hoped to be answered. Is this also applicable for research that is outside of biology?

[@danielm9463]

Yes! Inferential statistics is used across an extremely wide range of subjects. But this is very much an introductory video, so it's just important to know that the deeper you go into a subject, the more specific and advanced the statistics will generally become.

[@mellizacosminatan7173]

Thank you so much, this is really helpful. I am even able to understand my own research thoroughly without thinking what was the reason why is it not relationship based.

[@whisperingflute7104]

Which test is used to know if there's any relationship between two categorical variables?

[@danielm9463]

Hard to say without details, but it sounds like chi squared test of independence. Check out this video: kzbin.info/www/bejne/p4qyfoJuaL9kjbc

[@whisperingflute7104]

@@danielm9463 okay thank you so much!

[@madipsychomadi3144]

I have a question please. Which test for a VI has two modalities "2 groups", and VD and a VARIABLE mediator?

[@danielm9463]

So the independent variable is categorical, and there's a mediator variable. Are the mediator variable and the dependent variable both continuous?

[@madipsychomadi3144]

@@danielm9463 VM interval and VD ordinal . thank you

[@danielm9463]

@@madipsychomadi3144 This is a much more complicated picture. My familiarity with mediator variables is only with scenarios where all three are continuous variables (the IV, MV, and DV). Your interval variable will be treated as continuous. An ordinal variable would usually be treated as categorical, but this is something you'll have to decide on. As I understand, if the three variables were all continuous, then you would regress the DV on the IV, then regress the MV on the IV, and then regress the DV on both the MV and the IV. In the last step, you expect to see the coefficient of the IV drop dramatically compared to the IV coefficient acquired in the first regression. The purpose of this exercise is to confirm that mediator variable does, in fact, mediate the relationship between the IV and DV. If you don't have continuous variables across the board, then I'm not sure what the approach would be.

[@danielm9463]

@@madipsychomadi3144 I was thinking more about this, and the correct approach may be to proceed with the 3 regressions, but code dummy variables for your IV and DV. This isn't exactly your situation, but something along these lines: www.researchgate.net/post/How_do_I_handle_a_mediation_analysis_if_my_independent_variable_is_categorical_and_my_dependent_variable_is_dichotomous

[@madipsychomadi3144]

@@danielm9463 I used Baron Kenny's regression model. We'll see how my teacher corrects it.

[@magdalenatobar1420]

Is carbohidrate content continuous or qualitative? Im investogating the effect of carbohidrate percentage on the liberation of CO2 during yeast fermentation and dont know what test I can apply. The prodution of CO2 is presented in a time v/s producction graph. Can anyone help me ?

[@danielm9463]

This is a little trickier, because each trial produces a separate graph. In order to compare the different graphs, you'll have to figure out how you will distill each graph into a single value. For example, you could compute the average slope from each graph and compare those production *rates,* or you could set a fixed amount of time and compare the *total CO2 mass* during a constant ∆t. There are probably other ways to do it too. For example, maybe CO2 production follows a characteristic graph where it increases steadily, and then it flattens out. If this were true, then perhaps it would be meaningful to measure the amount of time ∆t before to get to the flat part of the graph. Or maybe you have experimental constraints that make certain data collection processes easier than others--for example, perhaps it's easier to measure the total time ∆t to achieve a fixed amount of CO2 production. So there are many different ways to extract a single statistic/value from each graph. The "correct" way to analyze those graphs will depend on the specific scenario you're dealing with, and I'm not as knowledgeable about the biology part of it. But there's something important to keep in mind: the way you process the data will slightly change your research question, and you may need to adjust your research question to reflect this. For example, if you choose to compare CO2 production rates (in grams/second) for different carbohydrate percentages, then maybe your research question should refer to the *rate* of carbon dioxide production. You want to make sure the question is phrased to match your analysis. Otherwise, your analysis is answering a different question. Ultimately, your dependent variable is most likely going to be numerical/continuous, because it will probably be something like CO2 production rate (e.g. in grams/second), or total mass (in grams), or total time (in seconds), etc. And your dependent variable is ultimately carbohydrate percentage, as you've said. So if you run 10 trials with 10 different carbohydrate percentages, then you're going to have 10 graphs, which you perhaps convert into 10 different rates. Then, you could produce a single final graph showing CO2 production rate against carbohydrate percentage. The point I wanted to make is that you'll probably have e.g. 10 raw data graphs, and then 1 processed data graph (or maybe more, depending on how advanced your analysis gets).

[@alziraangeli9707]

Awesome video! Thank you so much, Professor! Cheers from Brazil!

[@BiotechArab]

Bravo! It is just the thing I needed to refresh my background after a long break

[@whiterose_chani]

Can you help me I'm still very confuse. I need to compare between 5 categories of weight of oil palm and number of fruitlet ( for fertile, non-fertile and parthenocarpy) what type of data analysis should i do?

[@danielm9463]

What is your outcome variable? On what quantity are you measuring and then comparing these 5 categories? If it's a numeric outcome (which would presumably be the case), then you'd be comparing a continuous outcome/y for discrete groups (categorical x). Which test has categorical x and continuous y? If your experiment involves counting the frequency with which some event occurs, then you have frequency data. This is essentially a categorical y (with 2 levels: event occurred vs. event didn't occur) and a categorical x (with 5 levels of weight/fruitlet combos). In that case, which test is used for categorical x and categorical y?

[@VikramKumar-yc5rl]

I have 3 independent groups (standard drug, new drug, combination of both) to test a eye drop drug trial. I need to test the significance of new drug for various symptoms( itching, tearing, burning sensation, etc; total 10 symptoms ). I have before and after data for each symptom. Sample size for each group is 13,14 and 15. Which test should I use? To choose test do you I also need to know the normality of data distribution?

[@danielm9463]

Sounds like you have paired data (dependent samples). Your independent variable is categorical, and you're trying to compare the groups on *continuous* outcome/dependent variables. You want to test if the difference between paired samples is 0. Given that, which test would you choose? You can do a test for normality (or sometimes it's sufficient to just look at the q-q plot). If you can't satisfy the requirement of normality, you want to turn to a nonparametric test.

[@dr.ujjvalrana5261]

Thorugh and simple to understand. Thanks

[@emeraldwei6672]

Hi! I am wondering the following case: if I have a sample population, and I want to test the difference between the level of interest on a film between the entire sample population and the people younger than 18 in the sample; is this a 1-sample or 2-sample t-test? If it's 2-sample? Many thanks

[@danielm9463]

Let's use X for your independent variable. Rather than comparing X = 'under 18 years old' with X = 'all ages', a more straightforward approach would be to compare X = 'child (

[@rochetulp9780]

Quick question. Can i also use the Chi-Square test for relationship between 2 categorical variables?

[@danielm9463]

Yep, that's what you'd use to compare frequency data / proportions for 2 categorical variables.

[@marwaa.6759]

simply perfect and clear explanation . Thanks for your great effort

[@sallsall3]

I also use chi square test for my accounting research bcz it has nominal and ordinal scale Not just for biology subject

[@zucledesma]

Very clear explanation. Thank you so much.

[@danielm9463]

Glad it helped!

[@marilinserrano7719]

hi there, quick help here. What if i want to measure the correlation between 2 categorical variables? Is it still possible to still use a correlation test?

[@danielm9463]

I think you want to use the word "association" rather than "correlation." Association is a broader term that encompasses correlation. If we're looking at the *linear* association between 2 *continuous* variables, then we're dealing with correlation (e.g. Pearson's correlation coefficient). If we're looking at the association between 2 *categorical variables* to see if they're independent, then I think that would be a Chi Squared test of independence.

[@shahidbud3862]

Comprehensive and over simplified.A great video.

[@tivishasaravanan6770]

Thank you for creating this useful video! I've actually been struggling with doing a statistical test for my Biology IA. If I'm looking at the effect of a chemical on germination duration of a seed, so relationship and continuous data, would you recommend doing a t-test or Anova for this as well, or is looking at pearson correlation and regression sufficient? Also, if the data is non-linear (it actually decreases up to a certain point at the optimum, then increases from there), which statistical test would be most beneficial?

[@danielm9463]

It definitely sounds like your Y variable ("germination duration") is continuous, as you've said. But what would you say your X / independent variable is? Once you define it, would you say that it is continuous or discrete? I'm not sure correlation and regression will be available to you. Typically, if you're able to do a regression/ correlation analysis, then you *aren't* really able to do a t-test/ANOVA. A t-test/ANOVA implies that the independent variable is categorical, whereas regression/correlation implies that the independent variable is continuous.

[@tivishasaravanan6770]

@@danielm9463 i actually changed the concentration of the chemical, so it is continuous right? But even if I use regression/correlation, it would not be very clear because of the turning point in the middle, so what shall I do then?

[@danielm9463]

​@@tivishasaravanan6770 Ah, got it. You're absolutely right that both variables are continuous. It's a really interesting scenario you've got. Here are some ways to think about it. 1. This isn't really emphasized in an IB course, but it's incorrect to use linear regression and Pearson's correlation when the data don't have a linear relationship. Your IB Bio teacher (and the IB examiner) may not know that, however, and you could probably get away with calculating Pearson's correlation, and adding a caveat that the caution is needed because data fail to meet the condition of linearity. 2. One possible way to proceed (which I think you were referring to) is to "bin" your data. So, this means you break your concentrations into meaningful groups (e.g., 1-5 mg, 6-10 mg, etc.), and calculate the mean Y values for each categorical X value. Then you can do a t-test (if there are 2 groups) or ANOVA (if there are more than 2). This isn't really the best approach, but if you went this route, then ideally you'd have some sort of biology-based justification for the particular groupings you choose. (You want to at least attempt to answer the question, "why did you use 5-mg intervals instead of 10-mg intervals?"). IB would probably be impressed with this approach, but it's somewhat backwards, because you're treating your data as nominal and ignoring the fact that the 6-10 mg group represents *more* than the 1-5 mg group. Leaving information like this out of our statistical test lowers our ability to draw conclusive inferences. 3. So, instead of binning your data and *ignoring* the inherent ordering, a better approach (which uses the order) is Spearman's rank correlation coefficient. For your scenario, this is basically a non-parametric version of Pearson's correlation coefficient. Under this approach, you could either (1) use the exact concentrations (this would be easiest and probably best), or (2) bin the data into groups like described above (although doing this sacrifices some of the information in your data). If you go this route, it's worth reading a little about Spearman's rank correlation coefficient, but essentially it tests for a "monotone" relationship. This just means it tests for whether y is strictly increasing or strictly decreasing with x. Linearity is one type of monotonic relationship. (In addition to y = ax + b, y = ax^2 + bx + c is also monotone, etc.) 4. If the kink makes it look like there are two different *linear* trends, then you can always fit two linear functions (one for the first part of the data, and another for the second part of the data). This could lead to some rich analysis, because it allows you to talk about the threshold concentration, the changes in slope from one segment to the other, etc. Going this route, you could even report Pearson's correlation for each data subset. But it probably entails more analysis, and ultimately, IB wants you to be writing a Bio IA (not a Statistics IA). So it's about finding a balance. Fortunately, I don't think you have to go too in-depth on your stats calculations, which would mean you can do a lot of analysis, justify which test you used, and then jump straight to a discussion of how it relates to your research question & conclusion etc. without spending too much time explaining the calculation. (I think this is correct--you should double-check this with your teacher.) 5. If you have two distinct trends, but neither is linear, you can break your data into 2 subsets and use Spearman's rank correlation coefficient on each subset. Or, you could look for nonlinear trends within each subset. The easiest way to look for non-linear trends is guess-and-check. (e.g., try plotting y vs. x^2, y vs. 1/x, y vs. sqrt[x], etc. and see if any produce a straight line. If you find that y vs. x^2 is linear, then run a regression on y vs. x^2. Your regression line will have the form y = ax^2 +b, and the regression output will give you estimates of a and b.) The way to think about statistical tests is: (a) do the data satisfy the assumptions, and (b) if I treat the data differently (e.g. treat continuous data as though it's ordinal data as Spearman's correlation coefficient does), what information am I leaving out of my statistical test? Am I okay with leaving out that information? These are discussions you can have in your report. In general, the more information you leave out, the less powerful the statistical test becomes at finding a statistically significant effect. It's completely possible that you get a high p-value even when a true effect exists. That scenario arises when the test isn't powerful enough to detect the true effect, or the data have too much variance to detect the signal from the noise. But at the end of the day, what IB wants is for you to make decisions, justify them, and discuss the consequences. The more you can tie the statistical analysis back to your research question and your biology context, the better off you'll be.

[@gutsandglory7507]

@@danielm9463 I am impressed

[@PriitPauklin]

Great job! It helped a lot in explaining the basics! Nice work!

[@dirkaloha]

Reaffirmed what I am beginning to grasp. Thank you!

[@eslavathpuramdas8107]

Which test will be employed for comparison of dissolution profile between two products having continuous variable. Thanks for your time.

[@danielm9463]

Do you want to compare two curves/functions? This would be a more complex procedure, and the appropriate technique will be guided by conventions specific to industry/domain. For example, you might read through this: blog.eglifesciences.com/dissolution-analyses-comparison-of-profiles-using-f2-analysis-calculation Techniques like the Kolmogorov-Smirnov test might be available to you, or if a single defining quantity can be extracted from each dissolution curve (with standard error), you could potentially do a comparison that way.

[@houssemsama1535]

great video i have a question : i have a question : if i want to compare the sensitivity of 2 reagents which method should i use?

[@albertsays2678]

Cant correlation analyses deal with both continuous and categorical data? Cant we say that some categorical data have relationship to some continuousa data?

[@danielm9463]

Absolutely. That sort of analysis doesn't happen in an IB Biology classroom, but in other contexts of applied statistics, what you describe is very common. It's possible for us to do a linear regression using categorical variables that we encode using 'dummy variables' that take on a value of 0 or 1. In multivariate contexts, we can even test for interactions between the variables by multiplying a categorical dummy variable by another independent variable that is continuous.

[@danielm9463]

And in many contexts where we do a linear regression, we want to look at how well the model fits the data. So we looks at R^2 values, etc. to evaluate goodness of fit.

[@maryammagdelina]

Request advice/help : My study is musculoskeletal disorder among workers and the control measure taken to reduce the disorder. I have 1 (IV) production workers , my methodology for the study are using questionnaire and observations. I would like to relate the control measure I made to production workers in order to reduce the disorder. I’m still confused which statistical analysis I should use. Please advice me on this. 😭

[@yumi-yw6fu]

so if i have categorical data my research us qualitative?

[@danielm9463]

Good question! The answer is "it depends." One correction: you asked about "categorical data," but instead we should say "categorical variable." The variable is categorical vs. continuous. The data are quantitative vs. qualitative. This video didn't talk about qualitative data and was only focused on quantitative data. Categorical variables can produce quantitative and qualitative data. So can continuous variables--they can generate qualitative and quantitative data. Some examples of this are below. EXAMPLE 1 (categorical variables): Research question: do children prefer pizza over salad at a higher rate than adults? I will stand on the street and random ask people who walk by what they prefer. Quantitative data collected: Pizza Salad Child 24 13 Adult 6 17 Qualitative data collected: Children: they asked more questions. Several children what type of pizza and what type of salad. Adults: they answered more quickly and didn't have conversations about the topic. They seemed to take the question seriously. EXAMPLE 2 (continuous variables): Research question: does height increase linearly with age during childhood? I will pick one person at each age 2, 4, 6, ... 18 and measure their height with a tape measure. Quantitative data collected: Age Height (inches) 2 35 4 39 6 43 8 52 10 54 12 58 14 63 16 62 18 64 Qualitative data collected: Some of the younger children had a difficult time standing up straight. The child aged 2 and the child aged 4 were from the same family. All participants were born in Canada.

[@joyofall30]

Just starting to watch the video and loving your video

[@NinjaCoderInTraining]

Thanks man. Best explanation I've seen so far 👍

[@impossiblesoul]

which family would a test which has the purpose as relationship but have both categorical and continuous data go into ?

[@danielm9463]

Can you give an example? It would depend on which of the independent and dependent variables are categorical vs. continuous. The answer also depends on the level you're working at (e.g., high school, undergraduate/ university, graduate). There's a lot of simplification in this video, and the amount of detail quickly grows for students doing higher-level statistics. For example, if you have a categorical dependent variable, some categorical independent variables, and some continuous independent variables, then you use logistic regression (for a binary dependent variable) or multinomial regression (for a dependent variable that can take more than 2 levels/values) to evaluate the relationships of the predictors on the outcome. Similarly, if you have a continuous dependent variable, some categorical independent variables, and some continuous independent variables, then you can use multiple linear regression (as well as generalized linear models, e.g., you can use a Poisson regression for count data) to evaluate the relationships of the predictors on the outcome.

[@impossiblesoul]

@@danielm9463 for example i have a question which asks which test to use if we are trying to understand the relationship between car speed and stopping distance using 3 makes of tyre. i thought you can use a two way anova but i keep getting confused because we have more than two variables.. the way the question is worded makes me think that the tyre make is the response variable but in that case wouldn’t we use a linear regression … (so the car speed increases at increments , the stopping distance is measured , and then the car tyre type is changed and the experiment repeated)

[@danielm9463]

@@impossiblesoul There are a couple ways you could go with this. Generally, though, I think the question is asking: how does [the tyre make] impact [the relationship between car speed and stopping distance]? Are you doing this in a physics class, and if so, have you learned what physical quantity is responsible for one tyre having longer vs. shorter stopping distance? The force that brings the car to rest is the force of kinetic friction (also called dynamic friction), but that's an extrinsic property that depends on how hard you push the breaks. There's a different quantity that is intrinsic to the tyre material (and the material you're driving on), which relates to friction. If everything else is the same (same car mass, same initial velocity, final velocity = 0, and g=9.8m/s^2), then can you derive an expression between the stopping distance s and the quantity? As a hint, you would need Newton's second law and this kinematic equation: v^2 = u^2 + 2(a)(s). But back to your question: there is a numeric quantity that could replace [the relationship between initial car speed and stopping distance]. Then, the question would be asking: "how does [the tyre make] impact [this numeric quantity]?" And now you have a categorical variable (tyre make) impacting a continuous variable (the numeric quantity you may have learned about which pertains to friction). If you're doing this in a higher-level stats course, and not a physics course, then you have two independent/predictor variables, and they interact (the effect of speed on stopping distance changes based on the tyre make). In this case, the question is asking you to quantify the interaction between tyre make and speed.

[@impossiblesoul]

@@danielm9463 Thank you for the reply. I am doing a stats course and this question is more of us trying to understand what test can be used in this situation.

[@impossiblesoul]

Using what you said with there being two independent variables, I am going to with a two way ANOVA test

[@Melissa_Official454]

One of the best explanations. Thanks much.

[@rekharana6562]

Although the video is in English and my native language is Hindi, but I found it very much clear and informative.... thank you so much

[@erwincarlogonzales]

This video is golden for MS and PhD students

[@danielm9463]

I'm glad it's helpful! If you're an MS or PhD student, it could be useful for big-picture frameworks. But just a word of caution--it does make some simplifications. I've gotten comments from grad school professors that the distinctions here are less valid at a more advanced level, and that this video ignores e.g. generalized linear models, which extend regression to categorical variables. Those things are all true, but they're too advanced for an introductory level. I made this video for high school students with no background in statistics who have the task of determining an appropriate statistical test to use for their research project.

[@roopalivs6782]

Beautifully explained and simplified...... keep going 👍🏾

[@rediettekle7377]

Wow this simplified everything so much, thank you

[@mathildearctander1282]

How do you do anova when there is 9 independent variables and 1 dependent variable in the analysis?

[@danielm9463]

Let's say you're looking at factors that contribute to diabetes. The dependent variable is blood glucose level. Are there 9 different independent variables, or is there just a single independent variable with 9 different levels? If there are 9 different IV's they could be, for example, 1. age, 2. height, 3. weight, 4. sex, 5. race, 6. ethnicity, 7. history of heart disease, 8. total cholesterol level, and 9. daily sugar intake. On the other hand, if you have 1 independent variable (like "age group") with 9 levels (0-10 years old, 11-20 years old, 21-30 years old, 31-40, 41-50, 51-60, 61-70, 71-80, and 81-90).

[@zaina9011]

Best ever teaching technique....

[@mylearning7504]

Best video on statistical tests

[@zihao8027]

tysm!! i would've had a breakdown if not for your video 🤧 also, the dun dun dun at 0:08 made me laugh so thanks for that too

[@maryannmagtabog7284]

What if I have 1 independent seeking for the relationship and equation of 2 or more dependent variable? I know it is part of regression but I don't know which one, since multiple regression applies if I have 2 or more independent and only 1 dependent variable. Please enlighten me with this. Thank you!

[@danielm9463]

It gets a bit technical and starts to depend on what level you're working at. The simplest approach is to just do a different regression for each of your dependent variables. (This would be fine for a high school IB Biology IA.) However, running different regressions wouldn't give an accurate picture of your standard errors and confidence intervals. So if you're at a college/ graduate school level, you'd probably want to do multivariate regression (data.library.virginia.edu/getting-started-with-multivariate-multiple-regression/). This all assumes that your independent & dependent variables continuous (or, if you have categorical independent variables, then you're coding them with dummy/ indicator variables).

[@maryannmagtabog7284]

@@danielm9463 this is noted. thank you so much!

[@DrAamir580]

Sir you just nailed this video, you made me so damn clear about correlation. Thanks alot for your precious video😍

[@asmasultana2732]

Hello. I am a student who needs to work on the Stata assignment. I am facing difficulty, can I get some help and guidance.