Ayúdanos a dar forma a GoConqr. ¿Qué deberíamos añadir?

Apuntes

Explanation of Statistical Tests and Null/Alternative Hypotheses

41elocnehpets Inglés 1 página

Contenido del apunte (1)

Página 1

Null and alternative hypotheses

There is a great deal of uncertainty when carrying out experiments in biology due to natural variation between individuals. When assessing the results of an experiment we can never be 100% sure whether what we are seeing is significant and has an underlying biological reason or if it is purely due to chance.

The purpose of hypothesis testing in biology is to allow us to determine whether or not we can be highly confident that our results our significant and not due to natural variation. We form a null hypothesis which effectively states that nothing significant is occurring (that there is no difference between two mean values, no correlation between two variables, or no deviation from an assumed model); any trends or deviations are simply due to chance. The alternative hypothesis states that the null hypothesis is false. We only reject the null hypothesis if the probability of the results from our experiment happening due to chance is below a certain level; this is known as the significance level and in biology we use a significance level of 5%.

The process of hypothesis testing is in many ways similar to a criminal trial. Imagine we have a man on trial for murder. Then we wish to test the hypothesis that the man committed the murder. The defendant is pleading innocent: this plea of innocence is the null hypothesis. The alternative hypothesis is that he is guilty. The policy in a criminal trial is that the defendant is "innocent until proven guilty": that is, we assume our null hypothesis to be true unless the evidence allows us to believe, beyond reasonable doubt, that the defendant is in fact guilty.

There is a collection of evidence suggesting that the man did, or did not, commit the murder, in the form of fingerprints, witness statements, CCTV footage etc. During the trial, the jury will assess this evidence in order to make a judgement. This is analogous to us assessing the data from a biological experiment and carrying out a statistical test on it.

Finally, the jury must come to a verdict. If, from the evidence presented, they can be more than 95% sure that the man committed the murder (i.e. the probability of the presence of fingerprints, and other forensic evidence, being due to chance is less than 5%) then they will reject his plea of innocence (the null hypothesis) and find him guilty.

If the jury cannot be more than 95% sure of his guilt, then they will accept the null hypothesis and find him innocent. They believe that the probability of the evidence arising due to chance is greater than 5%. Note that this is not necessarily the same as the jury believing the man to be innocent - they just aren't confident enough to convict him.

Now let's look at the three types of statistical test you need to be able to carry out for A2 Biology, and when to use each one.

Standard error and 95% confidence limits

You will use this statistical test if you are comparing the means of two sets of data to determine if there is a difference between the two groups. The data in each set will be collected by taking measurements from a sample of a population, or by carrying out an experimental procedure that you repeat several times. For example, you might want to test the hypothesis the boys in sixth form are, on average, taller than the girls. Let's say there are 100 boys and 100 girls in sixth form. It would take a long time to measure the height of all 200 students and then calculate a mean - instead, you would take a sample of, say, 10 boys and 10 girls, measure their heights and assume the mean of each sample to be representative of the mean of the population.

Variation introduces uncertainty when taking measurements from a sample rather than the entire population. Some students will be very tall and others will be very small, and you cannot be certain that your sample will be representative of the population. There are 17.3 trillion possible samples of 10 boys you could take out of the whole population of 100, and each of these samples would have a different mean! A statistic called standard error (calculated as the sample standard deviation divided by the square root of the sample size) tells us how variable the mean is and is thus a measure of variation within the population.

How does the standard error help us? Say we calculate the mean height of the 10 girls to be 171.4cm, and the mean of the boys to be 175.2cm. At first glance it would seem that we can immediately confirm our hypothesis - but there is a problem. Because this is only a sample, the actual mean of the 100 girls could be quite different - we could, by pure chance, have chosen 10 of the tallest girls in the sixth form! However, it turns out that, 95% of the time, the actual mean of the population will differ from the mean of the sample by no more than 2 standard errors. Let's say we calculated the sample standard deviation for the girls to be 4.8cm. Then the standard error would be 4.8/(sqrt10) =1.518cm. We can thus be 95% confident that the actual mean of the population lies between (171.4 - 2 x 1.518) = 168.4cm and (171.4 + 2 x 1.518) = 174.4cm. For this reason, we refer to these numbers as the 95% confidence limits.

We can do a similar calculation for the boys. Suppose the standard deviation for the boys was 5.2cm. Then the standard error is 1.644cm. We can be 95% confidence that the mean of the population lies between 171.9cm and 178.5cm.

So our 95% confidence limits are:

For the girls: 168.4cm - 174.4cm

For the boys: 171.9cm - 178.5cm

We can see that these intervals overlap, as the upper confidence limit for the girls is greater than the lower confidence limit for the boys. We can therefore not be more than 95% confident that the mean of the boys' heights is different from the mean of the girls' heights, so we accept our null hypothesis. If the intervals did not overlap, we would have rejected it.

On BIOL6 paper 1 you will always be asked the same questions about statistics, and you have to have certain words in your answer to get the marks, so you should learn stock answers to these questions. When doing a standard error and 95% confidence limits test you should answer the questions as follows:

Which statistical test will you use? Standard error and 95% confidence limits

Give a reason for your choice of statistical test. Comparing two means

Calculate the test statistic. You should be very clear in the presentation of your working out here. First calculate the mean and standard deviation for the first data set (in this case, the height of the boys). Make sure you know how to calculate the standard deviation of a set of data, although this rubbishy website won't let me put the equation in (no offence, ExamTime admins). Then divide the standard deviation by the square root of the sample size to find standard error. Next, find the upper confidence limit (mean + 2 x SE) and lower confidence limit (mean - 2 x SE).

Do the same calculation for the second data set. Then, to make your answer perfectly clear, write the two intervals beneath all your working, for instance:

Confidence limits for boys: 171.9cm - 178.5 cm

Confidence limits for girls: 168.4cm - 174.4 cm

You might then finish your answer with a simple statement of whether or not these overlap (though this is not essential).

Interpret your test statistic in relation to your null hypothesis. Use the words "probability" and "chance" in your answer.

This will go one of two ways, depending whether or not your intervals overlap:

The confidence limits overlap: I accept my null hypothesis as the probability that the results are due to chance is greater than 5%.

The confidence limits do not overlap: I reject my null hypothesis as the probability that the results are due to chance is less than 5%.

Chi-Squared (Goodness of Fit)

Mathematical models can be used in biology to predict the frequencies of individuals in different categories. A good example is the Hardy-Weinberg equilibrium. For a gene with two alleles, the Hardy-Weinberg equilibrium predicts that the genotype ratios (homozygous dominant: heterozygous: homozygous recessive) should remain constant from generation to generation, and hence so should the allele frequencies. We know that the Hardy-Weinberg equilibrium is subject certain conditions:

=> Natural selection is not acting on the population

=> Mating is random

=> There is no immigration or emigration from the population.

=> No new alleles arise by mutations.

If, therefore, we find that the Hardy-Weinberg equilibrium does not hold true for a population - and that the allele frequencies have changed - we can conclude that one or more of those conditions was not true. For example, if directional selection acts on a population, we know that the allele frequencies at one extreme will increase as individuals with these alleles are more likely to survive and reproduce.

Chi-squared is a statistic that quantifies the goodness of fit of a model. We use the model to generate a set of expected frequencies: how many individuals would we expect to find in each category if the model is perfect? The data collected give us a set of observed frequencies, which are likely to differ from the expected frequencies. Chi-squared is a measure of how much the observed frequencies deviate from the expected frequencies. Our null hypothesis is therefore that there is no difference between the observed and expected frequencies. The larger the value of chi-squared, the more the data deviate from the frequencies we would expect if the model were correct, and so the less likely it is that our model is a good one. There is a cut-off value called the critical value: if chi-squared lies above the critical value, there is a less than 5% chance of these deviations occurring due to chance; we can reject our null hypothesis and say with some confidence that something has happened to alter the frequencies (e.g. natural selection).

Let's take an example. Codominant alleles are good to use as an example here because we can tell the genotype of an individual from their phenotype. Snapdragons have two alleles for flower colour: one which codes for white flowers and one which codes for red flowers. Heterozygotes for the gene have pink flowers.

In 2010, scientists sampled a population of snapdragons in a woodland and counted 34 with red flowers, 52 with white flowers and 37 with pink flowers. In 2014, scientists took a sample from the same woodland. This time, there were 74 with red flowers, 76 with white flowers and 66 with pink flowers. We wish to test the hypothesis that the allele frequencies have changed - since the scientists only took samples from the populations, the difference could simply be due to chance in the sampling.

Our null hypothesis is that the allele frequencies have not changed, and hence that the Hardy-Weinberg model is met. Assuming this is true, we would expect the proportions of red, white and pink flowers in 2014 to be the same as they were in 2010. This tells us that our expected frequencies come from the proportions in 2010. The observed frequencies are those seen in 2014.

In 2010, a total of 123 snapdragons were sampled, so as a proportion, 34/123 had white flowers, 52/123 had red flowers and 37/123 had pink flowers. In 2014, 216 snapdragons were sampled, so we multiply each of these proportions by 216 to get the expected number of plants with each flower colour:

=> (34/123)*216=59.7=>60 plants should have red flowers.

=> (52/123)*216=91.3=>91 plants should have white flowers.

=> (37/123)*216=65.0=>65 plants should have pink flowers.

We are now ready to calculate chi-squared! For each category, we subtract the expected frequency from the observed frequency, square it, and then divide by the expected frequency. We then add the three numbers together to get chi-squared.

For red flowers: O=74; E=60. (74-60)^2/60 = 3.27

For white flowers: O=76; E=91. (76-91)^2/60=3.75

For pink flowers: O=66; E=65. (66-65)^2/60=0.02

Chi-squared=3.27+3.75+0.02=7.04

We then look at the table on the Statistics sheet to find the critical value. The degrees of freedom is simply one less than the number of categories - so our test has 2 degrees of freedom. The critical value for 2 degrees of freedom is 5.99. Our value of chi-squared exceeds the critical value, so we reject the null hypothesis: we can be very confident that the allele frequencies of the snapdragons in the woodland have changed, and therefore this population does not fit the Hardy-Weinberg model: one or more of its conditions must be violated. In particular, the observed frequency of red flowers is much higher than expected, whilst the frequency of white flowers is much lower. Perhaps the red flowers are better at attracting pollinators and thus have more reproductive success than those plain old white flowers.

So now to answer the paper 1 questions:

Which statistical test will you use? Chi-squared

Give a reason for your choice of statistical test. Comparing observed and expected frequencies

Calculate the test statistic. You should show the calculation of expected frequencies, and the value of (O-E)^2/E, for each category. Then add them up to find chi-squared. To top it off, you might want to write how many degrees of freedom you are using, the critical value, and a statement that your value is either less than, equal to or greater than the critical value (though, again, this is not essential).

Interpret the test statistic in relation to your null hypothesis. Use the words "probability" and "chance" in your answer. Again, this will depend on your test statistic.

Chi-squared is less than the critical value: I accept my null hypothesis as the probability that the results are due to chance is greater than 5%.

Chi-squared is greater than or equal to the critical value: I reject my null hypothesis as the probability that the results are due to chance is less than 5%.

Spearman Rank Correlation Coefficient

Have you trialed a fixed range of values (i.e. more than 2) of your independent variable, and for each one, measured some dependent variable? Are you looking to determine if there is some sort of trend between the independent and dependent variables? Then look no further than the Spearman Rank Correlation Coefficient!

The idea behind the Spearman Rank test is quite simple. For each value of the independent variable you tested, you have a corresponding value of the dependent variable. If there is a perfect positive correlation between your two variables, then when you rank the dependent and independent variables from highest to lowest, your pairs should end up with the same rankings. The bigger the difference between the rankings of your paired variables, the weaker the correlation (unless the differences are large enough that they represent a negative correlation. In A2 Biology you'll only do tests for variables that have a positive correlation).

Let's use photosynthesis as an example.

Información

Descripción
Apunte sobre Explanation of Statistical Tests and Null/Alternative Hypotheses, creado por 41elocnehpets el 20/01/2015.
Creado por
41elocnehpets · ene 2015
Actualizado
19 feb 2016
Visitas
34
Valoración

Escríbenos

¿Sobre qué es tu mensaje?

¿Qué deberíamos añadir?