1000 words essay

Summer 2014 – ENV1000 Assignment 1

  1  

Formulating and Testing Hypotheses in Environmental Science Due July 21st by 11:59 pm

According to Department of Geography policy, the penalty for late submission of

assignments is 10% per day. Late assignments will not be accepted more than 5 days past the due date without a petition for special consideration.

Introduction and Background: (If you have a strong background in science, you can probably skim this section quite a bit. However, if you are at all confused about hypothesis testing and the formulating of hypotheses, I hope you will find this helpful. Formulating hypotheses and developing study designs is a critical skill in environmental science, and will definitely show up on the Term Test and Final Exam.)

The assignment itself starts on Page 5. Please submit the answers through Blackboard

when the link becomes active. Hypothesis testing is the backbone of science. Hypotheses can be thought of as provisional statements that propose a possible explanation for a particular observed phenomenon. In environmental science, we propose hypotheses and test hypotheses to understand patterns and relationships in nature, and also to understand the effects and consequences of human activities on the environment (positive, negative and neutral). All of the facts and concepts presented in the lectures and in the textbooks exist as the result of hypothesis testing. These are the current ideas about causal relationships, patterns and processes driving all kinds of natural and anthropogenic phenomena. As I mentioned in Lecture 11, good hypotheses “breed” – good hypotheses should give rise to more hypotheses that allow us to further develop our understanding of the world around us – sometimes building on a foundation of “current knowledge’; sometimes subverting a currently held view through a paradigm shift (Darwin (and Wallace’s) theory of evolution via natural selection is a classic example of a paradigm shift in science). Understanding how hypotheses are formulated and tested – and being able to sniff out the difference between “good” hypotheses and “bad” hypotheses – is important to everyone. If you plan to go further in science, this will be your bread and butter. However, non-scientists need to know this as well, because all kinds of good decision- making depends on the ability to understand how studies are designed, carried out and interpreted. Consumer decisions, health-care decisions, voting decisions etc. should be based on facts – and we need to know if those facts are reliable (i.e., if they are based on solid evidence derived from properly conducted science). For example, the so-called “link” between autism and vaccines came from a study so spectacularly flawed on so many levels (indeed, it was found to be intentionally fraudulent and the person responsible was stripped of his medical licence) that it should never have found traction in the popular press. However, it did - and we are now paying the price in completely avoidable disease outbreaks. A bit of scientific literacy would have gone a long way! Formulating hypotheses: The human brain is wired to see patterns and draw inferences from them. Superstitions are a primitive form of inductive reasoning. We can work through a “silly” example to see how we move from observation to conclusion.

Summer 2014 – ENV1000 Assignment 1

  2  

Observation: I have noticed that whenever I wear a certain lucky tuque, the Montreal Canadiens tend to win. (YAY!) This pattern has led me to wonder – is there a causal relationship between my tuque and the success of my favourite team? (Like the beer commercial says, “It’s only weird if it doesn’t work!)

In other words, I am deducing a general rule from my observation – wearing my tuque should cause the Habs to win. So, using the hypothetico-deductive method (a fancy term for the scientific method) I will propose and test a hypothesis:

Hypothesis: “Wearing my lucky tuque on game day has a positive effect on the likelihood that the Habs will win.”

This is a provisional statement that proposes a possible explanation (wearing my tuque) for a particular observed phenomenon (the Habs win). **Note that my hypothesis doesn’t state that wearing my tuque means the Habs always win, because if I did, just one loss while I am wearing my tuque would falsify my hypothesis! If I am THAT confident that wearing my tuque would ALWAYS make the Habs win (in the same way that I am confident that if I jump in the air, I will land back on the ground), and had the evidence and theoretical basis to back it up – it wouldn’t be a hypothesis, it would be a THEORY (like evolution) or perhaps even a LAW (like gravity, only about tuques and hockey). Null hypothesis: You may have heard this term before; we don’t emphasize it a great deal in this course, but formally, a hypothesis is always paired with another option – the null. This can be thought of as an (often unstated) “OR” statement in relation to the hypothesis – it is a statement of “no effect” or “no relationship”. Because scientists by nature are a cautious lot, the “null” is considered the default (what we assume to be true unless there is evidence to the contrary), and the hypothesis (your provisional explanation) is considered the “alternative”. Thus we have the following pair:

“Wearing my lucky tuque on game day has a positive effect on the likelihood that the Habs will win (alternative hypothesis, or HA) OR “Wearing my lucky tuque on game

day has no effect on whether the Habs win or lose” (null hypothesis, or H0).

So in this case, the null hypothesis is not “When I don’t wear my tuque, the Habs lose” (which is really just another way of stating the HA) or “When I wear my tuque, the Habs lose” (which is the opposite idea, so presumably is not based on my initial observations). The null is the statement of “no effect”. Predictions: For our purposes we don’t need to be really rigorous about the distinction between hypotheses and predictions, but to be precise, a prediction is a statement of what you expect to happen if the premise (hypothesis) is true. Hypotheses and predictions can be thought of as “if…then..” statements: “IF wearing my tuque affects the outcome of the game (hypothesis), THEN when I wear it, the Habs should win more often than when I don’t wear it (prediction).” Which leads us to the next point…

Summer 2014 – ENV1000 Assignment 1

  3  

Study design: Predictions are a stepping-stone to your study design – they help you figure out how to test your hypothesis. The “if-then” statement above suggests that I need to compare the success of the Habs on days that I do and do not wear my tuque. So, my study design will be to randomly designate all 80 games in the season as being either “tuque” or “no tuque” games. I will control for other variables – I will stick to exactly the same pre-game ritual for every game; the only thing different will be whether I wear my tuque or not. I will use an appropriate statistical analysis to determine if the number of wins with tuque is significantly greater than the number of wins without tuque. I will repeat this for several seasons before I am ready to make a conclusion as to whether the null or alternate hypothesis is supported by my data. Then I will be ready to send my paper to peer review in the hopes of publishing my findings. “Good” hypotheses: In lecture 11, I talked a bit about “good” hypotheses, and made the point that a hypothesis doesn’t have to be “correct” (i.e., supported by the evidence) to be good – in fact, sometimes supporting the null hypothesis can be a really interesting and important finding. For example, you might remember a study recently that showed the mammograms given before the age of 50 had no effect on decreasing breast cancer mortality rates. In other words, the null hypothesis was supported, and this was a surprising and important finding. One measure of a good hypothesis is that it “breeds” – i.e., it gives rise to new hypotheses. If I find a link between my tuque-wearing and hockey wins, there are all kinds of further questions I could explore: “Is the effect different on home and away games?” “Is the effect more pronounced if I wear the tuque for 2 days?” “What if I don’t wash the tuque after a win?” “If I only wear it for half a game, does it tend to result in a tie?” And so on… A good hypothesis is sometimes described as being SMART – Specific, Measurable, Achievable, Realistic, and Time-bound. What do we mean by each of these criteria? Specific – is it defined well enough to be testable? Is it specific enough to be meaningful? For example, “Long-term exposure to airborne asbestos particles significantly increases the risk of developing mesothelioma compared to a non-exposed population” is a specific hypothesis that lays the groundwork for making a prediction, and for designing a study. “Asbestos causes cancer” is not specific and thus a poor hypothesis. Measurable – this encompasses the idea of “falsifiability” – is it possible to imagine conditions where the null could be supported? The hypothesis “Global temperatures would be lower if humans did not exist” is a bad hypothesis as formulated, because we cannot test it – we can’t time travel and set up an experiment without humans (who would monitor it? That’s kind of a freaky thought…). What we CAN do is try to look at how global temperatures have varied with the increase in carbon-generating human activities, and compare them to a theoretical baseline. Hypotheses that require time- travel or parallel universes are not testable and thus not “good” in the sense we are interested in. It's interesting, though, that mathematicians and physicists do these kinds of thought-experiments all the time – “What would happen if time ran backwards as well as forwards? What if the universe has 26 dimensions instead of 3?” In some cases, these ideas are explored entirely theoretically; in some cases, doing arcane things like smashing sub-atomic particles into each other indirectly tests these ideas. In any case, these are way beyond the scope of our course. For our purposes, a hypothesis needs to be testable to have value.

Summer 2014 – ENV1000 Assignment 1

  4  

Achievable – this is related to the concept of measurability, but in a more practical sense. Something could be measurable in theory, but not practically doable. If testing the hypothesis requires unreasonable amounts of monetary, natural or human resources, it’s probably not a very well formulated hypothesis. Another important point is whether the hypothesis can be tested ethically. Without getting into a lengthy discussion about research ethics, suffice it to say that a good hypothesis is one where the benefits of knowing the outcome of the hypothesis test outweighs the resource costs and ethical implications. Realistic – is the hypothesis grounded in some kind of existing framework of understanding, or is it kind of “out there”? Can we imagine a plausible mechanism for the relationship? Time-Bound – this can really be thought of as a dimension of “achievable” (but SMART is a better acronym than SMAR, I guess J). Can the hypothesis be tested in a realistic time frame? For example, if I applied my tuque hypothesis to the Habs’ success in the Stanley Cup final, I could only collect data in those years when they make it to the final (which, sadly, is not every year). I might not live long enough to get enough replicates for a reliable statistical test (sob). Better stick to regular season games. So, if we examine my tuque – hockey hypothesis, it does reasonably well on 4 out if 5 criteria: It is Specific, Measurable, Achievable and Time-Bound. However, it is not Realistic – as much as I fervently wish I COULD control the outcome of the games, I can’t think of a reasonable mechanism to explain how my tuque could have an effect. Thus, any statistical correlation between my tuque-wearing and hockey wins would probably be due to random chance. Too bad L. What makes a good study design? Sometimes, the study design is pretty much dictated by the hypothesis (for example, there really isn’t any other good way to test the tuque hypothesis other than comparing the win rate with and without the tuque), but other times there are many ways of designing a study. Remember the donut/cancer example in Lecture 11? There were many possible ways to test that hypothesis. Good designs tend to:

• have adequate samples sizes (it’s hard to know what is “adequate” without going into statistical theory, but suffice it to say, a sample size of 1 is not adequate!)

• be unbiased (for example, if you wanted to compare the effects of smoking on blood pressure, you wouldn’t compare the blood pressure of young, healthy smokers to aged, unfit non-smokers)

• be achievable (don’t require super-human effort or resources to collect the necessary data – for example, don’t require you to measure the size of every single fish in a lake)

• have some kind of comparison (to a relevant control group, to previous conditions, etc.)

• be ethical.

(This is a very brief overview – we could devote an entire course to study design! – but I hope that at least gives you a sense of a good design.) Ok, now that we understand about formulating hypotheses, designing studies, and weird superstitions, we can move on to the actual assignment!

Summer 2014 – ENV1000 Assignment 1

  5  

Just 3 days ago, the following press release made it into the news: (http://www.vancouverobserver.com/news/first-nations-cancer-linked-oil-sands-toxins- wild-food-study)

Here is an excerpt condensed from the above article: “Two northern Alberta First Nations downstream of massive oil sands smoke plumes and tailing ponds released a human health study Monday, implicating the growth of the industry to many serious Aboriginal health concerns, including cancer. The new scientific study states the region's “country food” contains elevated levels of toxic metals and carcinogens that members of the Mikisew Cree and Athabasca Chipewyan First Nations traditionally eat. The wild foods include: moose, ratroot, duck, wild mint, spruce gum, pickerel, caribou, and Labrador tea. Fish are no longer eaten from the Athabasca River, due to government health warnings. The study reveals these foods contained elevated levels of heavy metals and carcinogens, and that nearly a quarter of the Aboriginal participants — 23 out of 94 — had cancer, among other ailments. The push for the study was motivated by a deep distrust of provincial and federal health officials, who they say have “failed” to comprehensively study the issue, said the leaders. “One thing most striking… is that both province and federal governments refuse to do anything about [the high rates of cancer]. Even though the pressure is escalating,” said ACFN Chief Allan Adam. “We are being brainwashed by the Conservative government that everything is ok. It’s not,” he added.” There are many interesting issues here – including the distrust of Alberta-based researchers! -- but we will focus here on hypotheses and study designs. Although I think it’s great if you want to delve into this topic more deeply, you do not need to do any further background research to answer the following questions. (We will be talking about the tar sands a bit later in the course – stay tuned J). Q1. Which of these two options is the better hypothesis statement and why? (2 marks)

A) Exposure to the tar sands causes cancer. B) Exposure to tar-sand chemicals through the consumption of contaminated wildlife increases the risk of developing certain cancers.

Q2. Suppose we decided to test the hypothesis: “Exposure to chemicals from the tar sands via the consumption of contaminated fish increases the risk of developing

Summer 2014 – ENV1000 Assignment 1

  6  

cholangiocarcinoma (cancer of the bile duct)”. If this is the alternative hypothesis, what would be the null hypothesis? (1 mark) Q3. Suppose we decided to test the hypothesis: “Exposure to chemicals from the tar sands via the consumption of contaminated fish increases the risk of developing cholangiocarcinoma (cancer of the bile duct)”. What would be a reasonable prediction arising from this hypothesis? (Remember that a handy way to develop a prediction is to say “If the hypothesis is true, then…..” (2 marks) Q4. Suppose that the following study designs have been proposed to test the hypothesis “Exposure to chemicals from the tar sands through the consumption of contaminated fish increases the risk of developing cholangiocarcinoma (cancer of the bile duct)”. Very briefly critique each design: if you see a problem with it, say what you think the problem is. If you think it is fine as is, say so. Please be brief! Possible criticisms could be (but are not restricted to):

• not achievable • unethical • not measurable • no control group • results may not be reliable

• biased • too small a sample • does not test the hypothesis • there may be other untested

variables affecting the outcome

If you see several problems, mention up to 2 problems for a complete answer. 4a. “I would measure the concentration of certain tar-sand contaminants in the blood of a person with cancer and a person without cancer and see if they are different.” (1 mark) 4b. “I would interview 50 people living close to the tar sands who have this type of cancer and ask them to estimate how much wild fish they consumed weekly over the last 5 years.” (1 mark) 4c. “I would interview 50 people living close to the tar sands who have this type of cancer and ask them to estimate how much wild fish they consumed weekly over the last 5 years. Then, I would interview approximately the same number of people from Toronto who DON’T have this cancer and ask them how much wild fish they consumed weekly over the last 5 years and see if the amount of fish consumption differs between the 2 groups.” (1 mark) 4d. “I would randomly assign 100 people to 2 groups. One group would eat contaminated fish once a week for 5 years; the other group would eat uncontaminated fish once a week for 5 years. I would see if the rate of cholangiocarcinoma differs between the two groups.” (1 mark) 4e. “I would look at rates of cancer in the community and see if they started to rise after the tar sands development came in (compared to before the tar sands were developed).” (1 mark)

Summer 2014 – ENV1000 Assignment 1

  7  

4f. “If there are water quality records going back to before the tar sands development, I would see if changes in water quality (specifically, changes in levels of contaminants) are correlated with changes in cancer rates.” (1 mark) Total marks = 10; worth 4% of your final mark. (The actual link for submitting the assignment likely won’t be available until Monday July 14th, but you have a whole week after that to submit.)