24 | A2 PSYCHOLOGY: THE STUDENT’S TEXTBOOKsuch a case. You might employ an alpha like this,for instance, if you had been investigating howwell an experimental drug influenced the memoryof elderly people with Alzheimer’s disease. Inorder to put your drug onto the market <strong>and</strong> allowpeople to start taking it, you need to be very surethat it does what it is supposed to.If the p value you find does not fall at or belowyour chosen alpha, then you are unable to acceptyour hypothesis. Instead you are unable to rejectthe null hypothesis <strong>and</strong> you cannot say that yourresults were significant. In these circumstances,you have to say that your results were non-significant.This is quite different from insignificant bythe way! Your findings were not at all insignificant.Even a result that does not support the hypothesisis interesting to scientists, but statistically speaking,it is a non-significant result.InvestigationHypothesis <strong>and</strong>null hypothesisAlphaCalculatedp valueCorrect decisionIncorrectdecisionDoes eatingcrisps make youfeel sick?Hypothesis: Themore crisps you eatthe sicker you feel.Null hypothesis:You do not feelsicker by eatingmore crisps.0.05p≤0.02(p is less than orequal to 0.02)The p value is lessthan alpha.We can reject thenull hypothesis<strong>and</strong> say that theresult supports ourhypothesis.We can confidentlysat that we are95% certain thatour results did notoccur by chance.Our results aresignificant.A type 2 error:We cannot rejectthe null hypothesis<strong>and</strong> must reject thehypothesis.The implicationsof an error likethis may result inpeople eating loadsof crisps becausethey believe it willnot make them feelsick when in factit does.A type 1 error:We can rejectthe null hypothesis<strong>and</strong> accept thehypothesis.Does wearingperfume makefemales moreattractive tomales?Hypothesis: Wearingperfume makesfemales moreattractive to males.Null hypothesis:Wearing perfumedoes not makefemales moreattractive to males.0.05p≤0.09(p is less than orequal to 0.09)The p value is largerthan alpha.We cannot acceptour hypothesis <strong>and</strong>must retain our nullhypothesis.We can onlysay with 91%confidence thatour results did notoccur by chance.Our results arenon-significant.The implicationsof this error maybe serious for theperfume industry.Sales of perfumemay drop becausewomen may feelthat wearing itmakes no differenceto their perceivedattractiveness.Should weprescribe apotentially dangerousexperimentaldrugto those withschizophrenia?Hypothesis:Treatment withthe experimentaldrug improves thequality of life ofthose suffering withschizophrenia.Null hypothesis:Treatment with theexperimental drugdoes not improvethe quality of life ofthose suffering withschizophrenia.0.01p≤0.017(p is less than orequal to 0.017)The p value is lessthan alpha.We can reject thenull hypothesis<strong>and</strong> say that theresult supports ourhypothesis.We can coincidentallysay that weare 99% certainthat our resultsdid not occur bychance.Our results aresignificant at the0.01 level.A type 2 error:We cannot rejectthe null hypothesis<strong>and</strong> must reject thehypothesis.The implicationsof this error arepotentially veryserious. Millionswho suffer fromschizophrenia willnot benefit fromthis wonderful newdrug.Extract from A2 Psychology: The Student’s Textbook © <strong>Nigel</strong> <strong>Holt</strong> <strong>and</strong> <strong>Rob</strong> <strong>Lewis</strong> ISBN: 9781845901004 www.crownhouse.co.uk
DATA ANALYSIS AND REPORTING INVESTIGATIONS | 25Type 1 <strong>and</strong> Type 2 errorsAfter you have carried out some statistical tests,you must decide whether you accept or reject thenull hypothesis. There are two major errors thatcan be made when doing this. These are describedrather confusingly as type 1 <strong>and</strong> type 2 errors.Type 1 errorThis is also known as the ‘false positive’ error. It isthe mistake of rejecting the null hypothesis whenit is actually true. You have made the decision toaccept your hypothesis by mistake. A very strictalpha value makes this kind of mistake less likely.Type 2 errorThis is also known as the ‘false negative’ error.This is the mistake of accepting the null hypothesiswhen it is in fact false, <strong>and</strong> you should haverejected it in favour of your hypothesis. A verystrict alpha value makes this kind of mistake morelikely.Do not underestimate the importanceof learning about probability, significance,<strong>and</strong> type1/type 2 errors! Theyare a crucial aspect of data analysis <strong>and</strong>you can bet that they will feature frequentlyin exam questions!Choosing the correctstatistical testThe choice of your test is influenced by the type ofresearch you have done, <strong>and</strong> the type of measurementsyou have made.Levels of measurementThere are three types of measurements you canmake in psychological research. These are nominal,ordinal <strong>and</strong> interval. The best way to describeeach is with an example.Nominal levelIf you have nominal data you have data that canbe classified in categories. By this, we mean thatif something is in one category it cannot be inanother category also. For instance, if you arecounting up the number of men <strong>and</strong> women ata rugby match, you cannot have someone whocounts as a man <strong>and</strong> also as a woman. Similarly, ifyou make a trip to a safari park to carry out a surveyon animals you may want to count the numberof monkeys you see <strong>and</strong> the number of hippos.You cannot have a monkey that is also a hippo –they exist as discrete categories. If your data is likethis, it is described as nominal level data.Ordinal levelThe clue for this one is in the name. Ordinalsuggests that there is an order. Horse racing is agood example. Horses are recorded as finishingfirst, second, third, fourth <strong>and</strong> so on. The order inwhich they finish is the important thing, not thedistance between them. If your data is like this, insome kind of order or rank, then it is described asordinal level data.Interval levelIf you are measuring something on a scale,perhaps the height of something or the time ittakes someone to do something, then you areusing an interval scale. Time, temperature, weight<strong>and</strong> height are all examples of interval levels ofmeasurement.A knowledge of your experimental design <strong>and</strong>the level of measurement used will allow you toanswer three simple questions, which will lead youto the appropriate statistical test.Example 1Our research is concerned with the relationshipbetween a person’s level of sadness <strong>and</strong> theamount of chocolate they eat. Here we are measuringsadness ratings <strong>and</strong> volume of chocolateconsumed in 10 different people.Question 1: Do I have correlation data?The answer to this is YES. You are looking for arelationship between variables <strong>and</strong> so your choiceof test is the Spearman’s Rho.Example 2Our research is concerned with investigatingwhether males are better at science subjects thanfemales. We collect science test scores from 10males <strong>and</strong> 10 females <strong>and</strong> want to compare them.Question 1: Do I have correlation data?The answer here is NO. You are looking atdifferences.Question 2: Am I looking at numbers incategories?The answer is NO. You are looking at differences.Question 3: What type of design did I use?The answer to this is an independent samplesdesign. You have two groups, one male <strong>and</strong> onefemale. A comparison is made of the scores of differentparticipants in the different conditions. Thetest for you is the Mann-Whitney Test.Extract from A2 Psychology: The Student’s Textbook © <strong>Nigel</strong> <strong>Holt</strong> <strong>and</strong> <strong>Rob</strong> <strong>Lewis</strong> ISBN: 9781845901004 www.crownhouse.co.uk