# Chi-Squared tests (2): Chi-Squared tests (2): Use with

Chi-Squared tests (2): Chi-Squared tests (2): Use with nominal (categorical) data when all you have is the frequency with which certain events have occurred. categorical data (avoid this, where possible) "non-psycho" "psycho" score per participant (aim for this, where possible) The 2 Goodness of Goodness of Fit test: Compares an observed frequency distribution with an expected frequency distribution. No. of squirrels killed yearly on the A27 Useful when you have the observed frequencies for a number of mutuallyexclusive categories, and you want to decide if they have occurred equally frequently.

Number of dead squirrels 60 50 40 30 20 10 0 1 2 3 4 5 6 7 Year of study Observed frequency Expected frequency Which soap-powder name do shoppers like best? Each of 100 shoppers picks the powder name they like most.

Number of shoppers picking each name (observed frequencies): Washo Scruba Musty Stainzoff Beeo 40 35 5 10 10 Expected frequency for each category is total no.observations / number of categories 100 / 5 = 20. total 100 The formula for Chi-Square: 2 Washo 40 20 O: E: O E

2 E Scruba Musty 35 5 20 20 Stainzoff 10 20 Beeo total 10 100 20 100 (O-E): 20 15 -15 -10 -10

(O-E) 2 400 225 225 100 100 20 11.25 11.25 5 5 O E 2 E 2 = 52.5 Chi-squared is the sum of the squared

differences between each observed frequency and its associated expected frequency. The bigger the value of 2, the greater the difference between observed and expected frequencies. But how big does 2 have to be, to be regarded as Goodness of big? Is 52.5 Goodness of big? We compare our obtained 2 value to 2 values which would be obtained by chance. To do this, we need the Goodness of degrees of freedom: this is the number of categories (or Goodness of cells) minus one. We have a 2 value of 52.5, with 5-1 = 4 d.f. Tables show how likely various values of 2 are to occur by chance. e.g.: d.f. 1 2 3 4 5 probability level: .05 .01 .001 3.84 6.63 10.83 5.99 9.21 13.82

7.81 11.34 16.27 9.49 13.28 18.46 11.07 etc. etc. 52.5 is bigger than 18.46, a value of 2 which will occur by chance less than 1 times in a 1000 (p<.001). The sampling distribution of chi-square: Frequency with which 2 values occur purely by chance: With 4 d.f., 2 values of 9.49 or more are likely to occur by chance on less than .05 of occasions. Our obtained 2 = 52.5, with 4 d.f., p < .001. A 2 value this large is highly unlikely to have arisen by chance. It appears that the distribution of shoppers choices across soap-powder names is not random. Some names get picked more than we would expect by chance and some get picked less. The 2 test of association between two independent variables: Another common use of 2 is to determine whether there is an association between two independent variables. Is there an association between gender (male or

female: IV A) and soap powder (Washo, Musty, etc.: IV B)? This gives a 2 x 5 contingency table. Data for a random sample of 100 shoppers, 70 men and 30 women: Washoe Scrubbup Musty Stainoff Nogunge total male 10 12 5 3 40 70

female 6 2 1 20 1 30 totals: 16 14 6 23 41 100 To calculate expected frequencies: E = row total * column total

grand total Work out the expected frequency for each cell: Washoe Scrubbup male female totals: Musty Stainoff Nogunge total 70 10 12 5 3 40 (11.2) (9.8)

(4.2) (16.1) (28.7) 6 2 1 20 1 (4.8) (4.2) (1.8) (6.9) (12.3) 16 14 6

23 41 e.g. 11.2 = (16 * 70)/100 6.9 = (23 * 30)/100, etc. 30 100 Using exactly the same formula as before, we get 2 = 52.94. d.f. = (number of rows - 1) * (number of columns - 1). We have two rows and five columns, so d.f. = (2-1) * (5-1) = 4 d.f. Use the same table to assess the chances of obtaining a Chi-Squared value as large as this by chance; again p< .001. Conclusion: our observed frequencies are significantly different from the frequencies we would expect to obtain if there were no association between the two variables: i.e. the pattern of name preferences is different for men and women. Chi-Square test merely tells you that there is some

relationship (an association) between the two variables in question: it does not tell you anything about the causal relationship between the two variables. Here, it is reasonable to assume that gender causes people to pick different soap powder names; it's unlikely that soap powder names cause people to be male or female. However, in principle the direction of causality could equally well go in either direction. Assumptions of the Chi-Square test: 1. Observations must be independent: each subject must contribute to one and only one category. Otherwise the test results are completely invalid. 2. Problems arise when expected frequencies are very small. Chi-Square should not be used if more than 20% of the expected frequencies have a value of less than 5. (It does not matter what the observed frequencies are). Two solutions: combine some categories (if this is meaningful in your experiment), OR obtain more data (make the sample size bigger). 2 test of association - the one- d.f. case: Preferred TV programme: Stenders: Corrie: Row total:

Origin: North: 13 10 23 South: 5 24 29 18 34 52 Column total: With 1 d.f. (as with a 2 x 2 table), the obtained 2 value is inflated; some statisticians advocate using "Yates' Correction for Continuity" to make the 2 test more conservative (i.e. make 2 value smaller and hence less likely to be significant). Same procedure as before, except

(a) take the absolute value of O - E (i.e., ignore any negative signs). (b) Subtract 0.5 from each O-E, before squaring it. 2 O E 0.5 E Without Yates Correction: 2 = 8.74. With Yates Correction: 2 = 7.09. 2 Why you should avoid using Chi-Square if you can: Design studies so that you can avoid using ChiSquare! Frequency data give little information about participants' performance: all you have is knowledge about which category someone is in, a very crude measure. It's much more informative to obtain one or more scores per participant; scores give you more information about performance than categorical data (and can be used with better statistical tests). e.g. IQ: which is better - to know participants are Goodness of bright or Goodness of dim, or have their actual IQ scores? 2 Goodness of Fit test on "fast food" data, using SPSS:

Are all brands mentioned equally frequently? Analyze > Nonparametric Tests> Legacy Dialogs > Chi-Square Brand first mentioned Test Statistics Chi-Square a df Asymp. Sig. Brand first mentioned 1209.440 7 .000 a. 0 cells (.0%) have expected frequencies less than 5. The minimum expected cell frequency is 50.0. Burger King Domino Pizza KFC McDonalds Pizza Express Pizza Hut Wimpy Other Total Observed N 57

1 44 274 1 10 3 10 400 Expected N 50.0 50.0 50.0 50.0 50.0 50.0 50.0 50.0 Residual 7.0 -49.0 -6.0 224.0 -49.0 -40.0 -47.0 -40.0 2 test of association on "fast food" data, using SPSS: Is there an association between gender and brand first mentioned? Analyze > Descriptive Statistics > Crosstabs...

2 test of association on "fast food" data (continued): Is there an association between gender and brand first mentioned? Case Processing Summary 11 response categories gives too many expected frequencies < 5. Therefore confined analysis to Burger King, KFC and McDonalds. (Use "Select Cases" on "Data" menu to filter out unwanted response categories). Valid N Sex * Brand first mentioned Percent 375 Cases Missing N Percent 100.0% N

.0% Percent 375 100.0% Sex * Brand first mentioned Crosstabulation Sex Male Female Total Count Expected Count Count Expected Count Count Expected Count Brand first mentioned Burger King KFC McDonalds 30 21 135 28.3 21.8

135.9 27 23 139 28.7 22.2 138.1 57 44 274 57.0 44.0 274.0 Total 186 186.0 189 189.0 375 375.0 Chi-Square Tests Conclusion: no significant association between gender and brand first mentioned. (2 (2) = 0.28, p = .87)) 0 Total

Pearson Chi-Square Likelihood Ratio Linear-by-Linear Association N of Valid Cases Value .283a .283 .135 2 2 Asymp. Sig. (2-sided) .868 .868 1 .714 df 375 a. 0 cells (.0%) have expected count less than 5. The minimum expected count is 21.82.

## Recently Viewed Presentations

• Promoting the use of a self-management strategy among novice chiropractors treating individuals with spine pain: A mixed methods pilot quasi cluster-clinical trial. Owis Eilayyan, AlikiThomas, Sara Ahmed, Craig Jacobs, Anthony Tibbles, FadiAlzoubi, Andre Bussières
• What is ROPE? Research Opportunity and Performance Evidence. ARC and its assessors and the College of Expert members are aware that PI's will not be able to demonstrate ROPE akin to that of researchers or academic staff . There are...
• Tap into prior knowledge (what you already know!) Aesop's Fables. The Tortoise and the Hare. The Lion and the Mouse. The Ant and the Grasshopper
• LA DEUDA PÚBLICA DE BOLIVIA Con el co-auspicio de: BANCO MUNDIAL CONTENIDO Normativa Evaluación de la deuda Deuda y pobreza Fuentes de información de deuda 1.
• Post your question to Edmodo under "Lecture Attendance Question Week 1" Your answer should be approximately one paragraph in length. Author Microsoft Office User
• Submission of Certified Foreign Priority Documents and Translations Under the AIA . March 20, 2013. Foreign Priority Documents. Foreign Priority Documents- Pre-AIA ("Old") Certified Copy of Foreign Priority Document . Needed to be filed before patent issued.
• El único criterio es el número de documentos recogidos en las revistas indexadas en las bases de datos Thomson-ISI (1990-2005) Re-Ranking RI3. Diez países iberoamericanos. ... Sesgos de la bases WoK. Cobertura de la región insuficiente. Cobertura disciplinar (sociales, humanidades...