SPSS

POLS 3000 Computer Assignment #3

TASK 1:

• Review the Recode and Crosstabs videos. Again, these are examples and do not cover this exact assignment.

• Open up your ANES2016REVISED data file

• As before, WEIGHT by the variable PW2016_FULL

• As before, take the questions that ask about a respondent’s feelings about Hillary Clinton (V16) and Donald Trump (V17)

TASK 2:

New Procedure – RECODE

• Use variables V16 and V17 • RECODE both variables into just 3 categories: negative (<50), neutral (=50), positive (>50) – see • To make sure that you do not lose the original 0-100 variation in V16 and V17 RECODE each into a new variable (V16A and V17A) • Remember: you need to have two RECODE commands here, each ended by a period.

TASK 3:

New Procedure – CROSSTABS

• Produce a crosstabulation with the new variables V16A and V17A as your dependent variable (defining the rows) and

your previously chosen IV (V22A_30D, V42E_30D, or V47_30D) as your independent variable. • Ask for column percentages, lambda and chi-square. • Describe what each of these 3 figures tell you for both dependent variables (V16A and, separately, V17A)

Relevant Percentage Differences (descriptive):

Is your IV category 0 more/less positive/negative than your IV category 1 on their feelings towards Hillary Clinton? towards Donald Trump?

Lambda (descriptive): By what proportion do we reduce our error in guessing feelings with knowledge of your IV? Use the V16A and V17A ‘dependent’ values, not your IV Do you notice any interpretive differences between lambda and the relevant percentage differences?

Chi-square (inferential):

If different (%), are the differences (V16A, V17A) “statistically significant”? Can we confidently reject the possibility that, in the population from which this sample was drawn, your IV and feelings are “statistically independent” of each other?

Task 4:

• Transfer all relevant information to a WORD document with your commentary and answers to questions. • Include a reasonably thorough discussion of what you learned about the interaction between the

two dependent variables and your IV. Compare what you found using CROSSTABS with what you found using T-TESTS. Make the document look as professional as possible (another requirement in the job market).

Example with feeling thermometer-Obama (V15) as DV, V1 (binary gender) as IV—your variables will be different. After opening ANES2016REVISED.SAV with SPSS: Syntax (you may play around with the GUI menu but RECODES are easier with syntax. Note: you will need two RECODE commands—one for V16 (into V16A) and one for V17 (into V17A). Note: Do not use V1 (binary gender) as your IV. I’m using it only as an example. WEIGHT by PW2016_FULL. RECODE V15 (0 thru 49=1) (50=2) (51 thru 100=3) into V15A. CROSSTABS TABLES=v15A by V1/CELLS=COUNT COL/STATISTICS=lambda chisquare. The RECODE command reclassifies the Obama feeling thermometer (V15) responses into just three categories but preserves the original 0-100 coding (V15) by creating a new variable V15A. This is precautionary.

Relevant Output:

V15A * PRE FTF CASI / WEB: R self-identified gender Crosstabulation

PRE FTF CASI / WEB: R self-

identified gender

Total 0. Male 1. Female

V15A 1.00 Count 728 695 1423

% within PRE FTF CASI /

WEB: R self-identified

gender

42.6% 37.7% 40.1%

2.00 Count 71 93 164

% within PRE FTF CASI /

WEB: R self-identified

gender

4.2% 5.0% 4.6%

3.00 Count 909 1057 1966

% within PRE FTF CASI /

WEB: R self-identified

gender

53.2% 57.3% 55.3%

Total Count 1708 1845 3553

% within PRE FTF CASI /

WEB: R self-identified

gender

100.0% 100.0% 100.0%

Men were 4.9 percentage points (42.6-37.7) more likely to feel cool towards President Obama than females. They were 4.1 percentage points more likely than women to fell warm towards him.

Chi-Square Tests

Value df

Asymptotic

Significance (2-

sided)

Pearson Chi-Square 9.590a 2 .008

Likelihood Ratio 9.594 2 .008

Linear-by-Linear Association 7.759 1 .005

N of Valid Cases 3553 a. 0 cells (0.0%) have expected count less than 5. The minimum

expected count is 78.84.

Directional Measures

Value Asymptotic

Standard Errora Approximate Tb

Approximate

Significance

Nominal by Nominal Lambda Symmetric .010 .011 .875 .382

V15A Dependent .000 .000 .c .c

PRE FTF CASI / WEB: R self-

identified gender Dependent

.019 .022 .875 .382

Goodman and Kruskal tau V15A Dependent .002 .001 .001d PRE FTF CASI / WEB: R self-

identified gender Dependent

.003 .002 .008 d

a. Not assuming the null hypothesis.

b. Using the asymptotic standard error assuming the null hypothesis.

c. Cannot be computed because the asymptotic standard error equals zero.

d. Based on chi-square approximation

With 2 degrees of freedom, a chi-square of 9.590 is significant at below the .05 level. We can therefore confidently reject the possibility that, in the population from which this sample was drawn the proportional breakdown on the feeling thermometer is exactly the same for men and women. We can confidently reject statistical independence between gender and feelings in the population from which this sample was drawn.

With a lambda of 0, knowledge of gender does not proportionately reduce our error in guessing feelings towards Obama at all. Why is this information different from what we found with percentage differences and chi-square (hint: text, p. 200)?