**A Meta-Analysis of Gender Differences
in Consumer Behavior**

There is a great deal of knowledge about gender differences in
specific areas of values, attitudes, preferences, lifestyles and
behaviors. Seldom do we see these various areas being examined
simultaneously within a single study. The purpose of this note is to
present some results from studying the *TGI Mexico* database. This is
a survey of 11,400 persons between the ages of 12 and 64 years old conducted by *
Moctezuma y Asociados* during late 1999 and early 2000. The content of this
survey covers a large number of subject areas:

- Demographics (e.g. age, household composition, employment, education, socio-economic level, income, household products and services, etc)
- Newspaper readership (general readership of weekday and weekend newspapers)
- Broadcast television viewing
- Cable/satellite television viewing
- General television characteristics (e.g. daypart viewing, program preferences)
- Magazine readership (general readership as well as individual magazine titles)
- Radio listening (e.g. daypart listening, program preferences)
- Alternate media (internet, outdoor, cinema)
- Personal product consumption (e.g. health and beauty aids, beverages, snacks, apparel, travel, financial services, etc)
- Household product consumption (e.g. food, laundry detergent, infant care products, automobiles, electronics, etc)

We propose to run an extensive analysis of gender differences for all these variables within this single source database. We note that this is a carefully designed and executed survey sample that adheres to generally accepted principles of probability sample with several levels of independent quality control. By analyzing this one database, we have eliminated the temporal, coverage, design and implementation differences that would occur if we were trying to compare results coming from different studies.

Within the *TGI Mexico* database, there are over 42,000
data items. However, we regard some items as being more important than
others. For example, we regard an item such as "drank soft drink
within the last 7 days" as being more important than "drank Seven-Up
more than thirty days ago." For the purpose of this analysis, we have
restricted our analyses to 1,432 data items. Still, this is a project that
is bigger than any other study that we are aware of. These 1,432 data
items represents those variables that are most frequently used in demographic,
media and marketing applications.

Consider any of the 1,432 data items, say Activity X. Our analysis of gender difference takes the form of asking the question: Is the incidence of Activity X among males different from that among females? And we want to do this for each of the 1,432 data items. Here are our summary statistics --- the mean incidences across these 1,432 data items are 13.1% among males and 12.7% among females. This certainly does not appear to be a vast difference. However, averages may conceal underlying differences, and we will carefully unpeel what lies underneath.

We recognize that these incidences are derived from a survey sample, which is subject to sampling error in the sense that different samples may yield somewhat different answers. We take into account these sampling errors by applying a statistical test called the Student t-test on the difference in incidence between the two sexes. This yields a probability called the p-value, which represents the probability for obtaining a difference of the observed magnitude under the null hypothesis of no difference, and when this p-value is too low or too high, we reject the null hypothesis of no difference between the sexes.

We conduct the Student t-test separately for each of the 1,432 data items. Because we are going to analyze the results of these analyses, this is called a meta-analysis (see book references at the bottom of this page). We obtained a total of 1,432 p-values, of which 45% of them are greater than 0.975 and 17% are less than 0.025. If the incidences were in fact equal among the sexes in the population, we would have expected to see about 2.5% of the p-values greater than 0.975 and another 2.5% less than 0.025. Therefore, our observed p-values would cause us to believe strongly that some of these incidences are significantly different between the sexes.

So how do we reconcile the fact that the mean incidences are about the same between the sexes with a series of statistical tests that supports significant differences in incidences between the sexes? It is very simple --- men have higher incidences on certain activities, while women have higher incidences on other activities, but such that their mean incidences are close. For example, the mean magazine audience is about the same between males and females (1.8% vs 1.9%), but about one-third of the magazines have a predominantly male audience while another one-third of the magazines have a predominantly female audience.

In the following table, we show a detailed breakdown by subject area. The gender differences are seen to vary across the subject areas.

SUBJECT AREA |
# of activities | Mean incidence for males |
Mean incidence for females |
% p-value greater than 0.975 |
% p-value less than 0.025 |

TOTAL |
1432 | 13.1% | 12.7% | 45% | 17% |

Demographics | 210 | 18.1% | 16.5% | 57% | 20% |

Newspaper readership | 4 | 15.6% | 11.6% | 100% | 0% |

Broadcast TV viewing | 69 | 23.3% | 23.0% | 20% | 12% |

Cable TV viewing | 346 | 2.2% | 1.7% | 48% | 1% |

General TV characteristics | 239 | 14.7% | 14.4% | 46% | 26% |

Magazine readership | 107 | 1.8% | 1.9% | 30% | 33% |

Radio listening | 253 | 10.3% | 10.7% | 39% | 21% |

Alternate media | 31 | 11.2% | 8.6% | 84% | 3% |

Personal product usage | 98 | 26.9% | 27.6% | 55% | 29% |

Household product usage | 64 | 47.1% | 46.5% | 27% | 22% |

In interpreting these data, we should bear in mind that while they may indicate differences, they do not imply any judgment of superiority or inferiority. Within this broad set of differences between males and females, a data item has to be interpreted in light of its context. For example, if men are more likely to watch sports television program than women, then this is a piece of socio-cultural behavior that would be of commercial interest to television programmers, media planners and marketers. As another example, if men are more likely to have higher personal incomes than women, then this may be the result of inequities of the socio-economic system.

**BOOK REFERENCES ON META-ANALYSIS**

- Harris Cooper and Larry V. Hedges (1994)
**The Handbook of Research Synthesis**. Russell Sage Foundation: New York, NY - Gene V. Glass, Barry McGaw and Mary Lee Smith (1981)
**Meta Analysis in Social Research**. Sage Publications: Beverly Hills, Ca. - Larry V. Hedges and Ingram Olkin (1985).
**Statistical Models for Meta-Analysis**. Academic Press: New York. - John E. Hunter and Frank L. Schmidt (1990).
**Methods of Meta-Analysis: Correcting Error and Bias in Research Findings**. Sage Publications: Newbury Park, CA. - John E. Hunter, Frank L. Schmidt and Gregg B. Jackson
(1982)
**Meta-Analysis: Cumulating Research Findings Across Studies**. Sage Publications: Beverly Hills, CA. - Richard J. Light and David B. Pillemer (1984).
**Summing Up: The Science of Reviewing Research**. Harvard University Press: Cambridge, MA.

(posted by Roland Soong on 10/23/00)

(Return to **Zona Latina's Home Page**)