Methods of Communication Research and Statistics

The SummoScanner is a continuous survey of print media penetration. This exercise deals with a target group study conducted in 1998, for which a random sample of the Dutch population was taken of those aged 13 or older. In relation to a large number of newspapers and magazines, the respondents to the Summo surveys were asked to state 1) whether they ever read or browse through them, 2) whether they have read or browsed through the latest edition, and 3) how many of the past six editions they have read or browsed through. Based on the answers to the last two questions, a 'reading likelihood' is calculated (all variables with names beginning with 'lk' [lk = Reading Chance]). These variables are measured at interval level but often have a very skewed distribution; most respondents have a reading likelihood of zero. We are therefore going to continually dichotomise the reading likelihood: people who have a reading likelihood (greater than 0) and those who do not (0).

Database: Summo.Leeskans2.sav

a. Test the hypothesis that women who are likely to read Beau Monde are also likely to read Elegance. Based on the results, are you of the view that it is sufficient to place an advertisement in Beau Monde in order to reach current and potential readers of Elegance?

We always work with recoded versions of the reading likelihood variables, with one value being a reading likelihood of exactly 0, and the other a reading likelihood of greater than 0. We therefore have categorical variables and must carry out chi-square tests.

In exercise a, we effectively have to test whether there is an association between the recoded reading likelihoods for Beau Monde and Elegance. A contingency table with the chi-square value and a symmetrical measure of association (Phi or Cramer's V) will suffice. Don't forget to first select only the women!

There appears to be a significant association between the two reading likelihoods: Fisher's exact, p < 0.001. Note: we use the Fisher's exact test because we have a 2x2 contingency table.

Looking at the contingency table, we see that the observed frequency of respondents with a reading likelihood for both publications (f_o = 28) is greater than the expected value (f_e = 10.6, standardised cell residue = 5.3). This means that a reading likelihood for one publication does coincide relatively frequently with a reading likelihood for the other.

However, the association is weak (Phi = 0.19; p < 0.001). It is therefore definitely not the case that all or the majority of the respondents who are likely to read one publication are also likely to read the other. This means that you only reach a limited number of current or potential readers of one publication through the other. If there are funds available to place advertisements in both publications, this is recommended.

Dichotomous reading likelihood of Elegance * Dichotomous reading likelihood of Beau Monde Cross tabulation

Dichotomous reading likelihood of Beau Monde Total

No reading likelihood Reading likelihood

Dichotomous reading likelihood of Elegance No reading likelihood Count 789 37 826

Expected Count 771,6 54,4 826,0

Std. Residual ,6 -2,4

Reading likelihood Count 133 28 161

Expected Count 150,4 10,6 161,0

Std. Residual -1,4 5,3

Total Count 922 65 987

Expected Count 922,0 65,0 987,0

Chi-Square Tests

Value df Asymp. Sig. (2-sided) Exact Sig. (2-sided) Exact Sig. (1-sided)

Pearson Chi-Square 36,514(b) 1 ,000

Continuity Correction(a) 34,445 1 ,000

Likelihood Ratio 28,346 1 ,000

fisher's Exact Test

,000 ,000

Linear-by-Linear Association 36,477 1 ,000

N of Valid Cases 987

a Computed only for a 2x2 table
b 0 cells (,0%) have expected count less than 5. The minimum expected count is 10,60.

Symmetric Measures

Value Approx. Sig.

Nominal by Nominal Phi ,192 ,000

Cramer's V ,192 ,000

N of Valid Cases 987

a Not assuming the null hypothesis.
b Using the asymptotic standard error assuming the null hypothesis.

* Syntax from exercise a.
* First check the reading likelihood variables.
FREQUENCIES
  VARIABLES=lkele lkbmo
  /ORDER= ANALYSIS.
* There are no strange values.
* Dichotomise reading likelihood.
RECODE
  lkele lkbmo
  (0=0) (ELSE=1) INTO llkeleRE lkbmoRE.
VARIABLE LABELS llkeleRE
  'Dichotomous reading likelihood of Elegance '/lkbmoRE' Dichotomous reading likelihood of Beau Monde'.
EXECUTE.
* Define Variable Properties.
* llkeleRE.
VALUES llkeleRE
  0 'No reading likelihood'
  1 'Reading likelihood'.
* lkbmoRE.
VALUES llkeleRE
  0 'No reading likelihood'
  1 'Reading likelihood'.
EXECUTE.
* Select women.
USE ALL.
COMPUTE filter_$=(gesl = 2).
VARIABLE LABEL filter_$ 'gesl = 2 (FILTER)'.
VALUE LABELS filter_$ 0 'Not Selected' 1 'Selected'.
FORMAT filter_$ (f1.0).
FILTER BY filter_$.
* Contingency table, reading likelihood of Elegance and Beau Monde.
CROSSTABS
  /TABLES=llkeleRE BY lkbmoRE
  /FORMAT= AVALUE TABLES
  /STATISTIC=CHISQ PHI
  /CELLS= COUNT EXPECTED SRESID
  /COUNT ROUND CELL.

b. Are Elegance and Beau Monde aimed at wealthier women? Use dichotomous reading likelihoods and the variable 'INKOM' (income) to distinguish between less wealthy (average or below average) and wealthier (above average) women.

In the case of Elegance, we find an only just significant difference in reading likelihood between wealthier and less wealthy women, Fisher's exact p = 0.050. If 0.05 is the maximum permissible likelihood for incorrectly rejecting the null hypothesis, then we can only just reject that null hypothesis in this case. If you would like to know exactly what it is, open the table in the table editor (double click on the table) and then double click on the number 0.050. You will then see the more precise excess likelihood: 0.0498026310202801.

However, the association is very weak (Goodman and Kruskal's tau = 0.005, p = 0.047). Women with an above average income have a slightly more frequent reading likelihood (fo = 64) than expected on the basis of chance (fe = 53.4), while the reverse is true in the case of less wealthy women.

We find no significant difference in reading likelihood of Elegance between the wealthier and less wealthy women, Fisher's exact p = 0.323. Here, we have to assume that there is no difference in the population.

* Syntax from exercise b.
* Recode the income categories.
RECODE
  inkom
  (3=2) (1 thru 2=1) (ELSE=SYSMIS) INTO inkomRE.
VARIABLE LABELS inkomRE 'Dichotomous income category'.
EXECUTE.
* Define Variable Properties.
* inkomRE.
VALUE LABELS inkomRE
  1 'average or less'
  2 'above average'.
EXECUTE.
* Contingency table for reading likelihood according to average or less/above average income.
CROSSTABS
  /TABLES=llkeleRE lkbmoRE BY inkomRE
  /FORMAT= AVALUE TABLES
  /STATISTIC=CHISQ PHI LAMBDA
  /CELLS= COUNT EXPECTED SRESID
  /COUNT ROUND CELL .

SPSS exercise 4.6