A Place at the Policy Table? Behavior Genetics and Estimates of Family Environmental Effects on IQ

A Place at the Policy Table? Behavior Genetics and Estimates of Family Environmental Effects on IQ

David C. Rowe (1997)

The knowledge generated in behavior genetic studies is not often made a part of social policy deliberations. The argument of this article is that behavior genetics belongs at the social policy table. Perhaps ironically, behavior genetics is one of the best methods for understanding environmental influences. Behavior genetic studies can reveal which traits are most influenced by shared environment and, thus, which are most malleable through changes in shared environments. The current consensus of behavior genetic studies is that IQ is not a particularly malleable trait, especially after childhood. Furthermore, for working-to middle-class families, the shared environmental effects on IQ in childhood seem to be temporary rather than lasting. Behavior genetics also can estimate genetic and family environmental components of racial differences in IQ because quantitative genetic models now permit the simultaneous analysis of group means and individual variation. Although not as directly relevant to policy as targeted research on specific policy options, behavior genetics clearly deserves representation.

Differential Responsiveness to Exposure

Two processes are at work in intellectual growth. One is an exposure process. Each exposure to a school lesson (or a life lesson outside of formal schooling) creates an opportunity to learn. Some people attribute individual differences in intelligence to the number of exposures or opportunities available to each child (Ceci, 1990). By a literal interpretation of “exposure theory,” a child who read 100 books would be twice as good verbally as one who read 50; one who read 400 would be eight times better, and so on.

The limitation of exposure theory is that children will gain unequally with each exposure (see Carroll, 1997). Francis Galton (1869/1962) made an analogy between intellectual and physical development. Weight training makes nearly everyone stronger. But, depending on variation in physiology, one man may build muscle rapidly, whereas another, under the same regimen of training, builds muscle more slowly. Furthermore, few men can build to anything like Mr. Universe’s impressive bulk; some muscles are not as capable of this kind of growth as others. Not only does growth occur at different rates; the maxima achievable may differ as well.

Just like electronic hardware in a computer, the brain supports intellectual growth through its capability to assimilate and retain something from each learning exposure. Different brain physiologies mean that learning occurs at different rates. As Gardner (1995) has observed, in no domain of human talent is practice alone sufficient:

When one revisits the psychological variable that has been most intensively studied, that of psychometric intelligence or g, one finds little evidence to suggest that sheer practice, whether deliberate or not, produces large ultimate differences in performance . those individuals who combine high psychometric intelligence in childhood with diligent practice in (and out of) school are more likely to become expert thinkers or scholars than those who can only practice (so-called overachievers) or those who do not practice at all (so-called under-achievers). (p. 802)

The heritability of IQ (Plomin & Petrill, 1997) implicates an influence on psychometric intelligence most removed from everyday classroom experience: the DNA code. The genes involved and biological pathways that work from them to the IQ phenotype are largely unknown, and they will remain unknown until further advances are made in molecular genetic studies of IQ. Nonetheless, a broad spectrum of studies now directly connect variation in brain biology to IQ variation. IQ is associated with variation in the anatomical structure of the brain, with blood flow in the cerebral cortex, with EEG potentials, and with other measures of brain physiology (Andreasen et al., 1993; Detterman, 1992; Haier et al., 1988; Willerman, Schultz, Rutledge, & Bigler, 1991).

Differences between bright and dull people also can be identified through elementary information-processing tasks. Although this field is far from settled, general agreement does exist that brighter individuals demonstrate faster reaction times on simple tasks, for example, when lifting a finger in response to the onset of lights, or when making other exceedingly simple cognitive “decisions” (Jensen, 1993). On an auditory inspection task (Deary, 1995), brighter people are better able to detect which of two briefly (i.e., . 1-.5 sec.) presented tones is the one higher in musical pitch. This task is so simple that mentally retarded individuals can complete it without mental strain, yet in normal populations the speed of doing so is remarkably strongly associated with IQ, Y = -.30 to -.40, as brighter people are able to distinguish tones presented more briefly. Brighter people also have greater capacity to juggle information about in short-term memory (Jensen, 1993). When given competing memory tasks, they better recall earlier presented lists of digits or letters. Speed, memory capacity, sensitivity to brief stimuli all suggest that brain “wetware” differs between bright and dull people.

In summary, the passive exposure theory of intelligence is an incomplete one. Differences in intelligence arise both from unequal exposures to cultural information and from unequal capitalization upon them for reasons that seem partly biological. An 11-year-old boy may learn the word “sequestration” from hearing his parents discussing the famous O.J. Simpson trial, from reading an article on it in the newspaper, or from The Simpsons, a cartoon series in which the character Homer wanted to be sequestered to enjoy a nice hotel room. A task for learners is to attend to each exposure so that word meaning may be extracted and integrated with existing knowledge. Even if dull children have had many exposures, they do not necessarily learn more because critical steps of extracting and retaining meaning may fail.

Shared Versus Nonshared Environmental Variation

Environmental variation can be divided into shared and nonshared parts. Shared environments are correlated perfectly across siblings in a family, or across parent-child pairs. Thus, shared environments induce behavioral resemblance. In contrast, the nonshared environment contributes only to family members’ dissimilarity; it also includes measurement error (which by definition cannot impose a correlation).

The shared environmental component of variation is sometimes misunderstood. It does not arise in a general cultural background accessible to all members of a culture. The United States has McDonald’s, the ABC and NBC television networks, baseball and football, malls and video arcades, and a long list of other cultural commonalities. None may contribute to shared variation, however, because each one is “constant” and so cannot create IQ variation.

The shared environmental concept refers (by definition) to influences at the level of the family unit – a shared influence is one shared by family members, but one that nonetheless differs from one family to another. For example, “hours of TV viewed” qualifies not because television is ubiquitous in American culture, but rather because in some families the TV is watched for hours on end, whereas in other families, it is stowed away in a closet.

Twin and adoption studies sample those rare pairs of siblings who are atypical in genetic relatedness (e.g., MZ twins, adoptees). Such rare pairs of individuals usually do not reside in one neighborhood; thus, the sibling pairs would typically attend different schools and have different friends. In these research designs, neighborhood and school-level influences related to IQ should contribute strongly to the “shared” variance estimate in a behavior genetic study. Similarly, parental vocabulary, number of books in the home, parental encouragement of achievement are all primarily family-level, shared influences. Parental years of education is also shared because it would be highly correlated across siblings in the same family. As such, the primary environmental variables named by socialization theory, as related to intellectual growth, should appear in the apportionment of environmental variance within the shared rather than the nonshared component.

Of course, the shared component of environmental variation is a mathematical idealization. If parents treat their children alike 75% of the time and treat them differently 25% of the time, then the treatment would operate to make siblings both alike and different; family environment would be both shared and nonshared. Genes also do a double duty – full siblings receive one half of the same (variable) genes and one half different ones – hence, similarity in, say, height between siblings may coexist with different eye colors, hair textures, or moods.

According to the family effects theory of environmental influence on IQ, however, the shared should dominate. Figure 1 pictures a shared effect (SE) for unrelated siblings reared together (e.g., step-siblings or siblings who are both adopted). SE captures the part of any environmental effect that is correlated across siblings. Suppose the regression (c in Figure 1) of shared environment on IQ were .70, then the expected sibling correlation, P1 with P2, would be this regression coefficient squared, c² = .49, or about 50. Shared effects could be weak either because few IQ-relevant environments correlated across siblings, or because they did not much affect IQ. For instance, if the regression coefficient dropped from .70 to .10, then the expected sibling correlation would be only .01.

A Place at the Policy Table - Behavior Genetics and Estimates of Family Environmental Effects on IQ - Figure 1

Shared effects also can be estimated from parent-child pairs. The latter is instated in one of two ways. First, it may be the effect of the environment of the child’s grandparents’ family – in which the parents themselves were raised – on the child’s own environment. Or, it may be an effect of the phenotype of the parent directly on that of the child. One can also imagine environmental effects that are shared by siblings, but not by parent-child, for example, attending a neighborhood school. Thus, these different estimates of shared effects need not agree exactly.

The reader should understand that this theoretical effect of all shared experiences would exceed that of any specific family environmental influence. For example, if social class created a correlation on IQ for siblings of .20, then the shared environmental estimate should be greater than this value because it includes this effect plus all others. The effect of shared environment is given the symbol c² (c for common). As with the heritability coefficient, it ranges between 0 and 1. Technically, it is the percentage of trait variance among individuals explained by shared environmental influences.

The shared environmental estimate also can be thought of as an “effect size.” The c² for siblings gives the amount of change in a trait that would be expected if siblings were swapped between families (as they are, literally, when adopted) For example, suppose that the correlation for genetically unrelated sibling pairs was .40, then 40% of the variation in IQ would be attributed to shared environmental effects. If children were swapped between families that differed by 1 SD on the shared environmental (latent) scale, then an improvement of 9.5 IQ points could be expected (square root of .40 X s², where s = 15); a 2 SD swap should produce a 19-point IQ increase.

The greater c², the more change that should follow from having children go from one set of family circumstances to another, or by environmentally making similar changes in their own families. This interpretation of c² led Jencks (1980) to observe that this number is truly the best environmental “effect size” from a behavior genetic study:

Many policy proposals consist, in essence, of providing all families with advantages currently enjoyed by the privileged. If e²c [c² here] is initially large for a given phenotype, successful efforts along this line can be expected to substantially reduce the total variance of the relevant phenotype and greatly improve the relative position of the disadvantaged. (P. 734)

IQ Malleability Limited by Shared Environment (c²)

How much can we boost IQ for children in working- to professional-class families? On the basis of c² estimates, not very much. McGue et al. (1993) reviewed studies of IQ from a life-span perspective. In childhood, the c² estimates are ample, although less than heritability estimates: c² for adoptive siblings = .32, N = 714 pairs; c² for adoptive parent-child = .24, 720 pairs. However, these environmental effects were not sustained for adolescents and young adults. When recomputed for late adolescent or young adult adoptive pairs, the c² estimates were reduced to nearly zero, mean c² = -.01. Similarly, in an analysis of published twin correlations, the estimate of shared environment also decreased from childhood to adulthood. In young children, the twin studies yielded greater shared environmental estimates than the adoptive studies, about .40 for children 4 to 6 years old. However, the estimate for twins in adulthood was zero. Consequently, MZ twins raised apart were nearly as similar as MZ twins raised together. These studies directly challenge the conventional wisdom that IQ is strongly shaped by rearing environments.

The idea of “niche picking” illuminates the low effect sizes for family-related environments. If in the long run “genes drive experience,” brighter children will seek out and respond to those experiences promoting more advanced intellectual growth, whereas duller ones avoid, or cannot assimilate, those same experiences. More advantageous home experiences may give some children an initial intellectual advantage, which would create a shared environmental effect in childhood that can be attributed to learning opportunities. These early gains, however, would be unsustainable past childhood, because intellectual growth becomes more and more dependent upon the child’s active involvement. As the rate of intellectual growth is more closely aligned with individuals’ genotypes, IQ becomes more heritable and decoupled from shared family circumstances.

The nonshared environments, of course, remain important. Parents do affect their children, but the direction of that “nudge” is often unpredictable. Encouraging one child to study hard may make that child get better grades, whereas a brother or sister may rebel against being “bossed” by the parents. Now, some systematic nonshared environmental influences have been proposed, for example, birth order or birth spacing. Such variables make little contribution to IQ variation, however (e.g., Rodgers & Rowe, 1985). Although new statistical methods can be used to identify nonshared effects (Rodgers, Rowe, & Li, 1994), for the most part, we do not understand them, and some specific ones (e.g., accidents of embryological development) may not be practically modifiable. In particular, I do not believe that rearing influences that have been rejected as strong shared effects will emerge as strong nonshared ones.

I should caution that IQ is a “phenotype” (i.e., observed trait), not a measure of “genetic intelligence.” To say that “IQ becomes more closely aligned with genotypes” means that a probabilistic inference from phenotype to genotype becomes stronger – but it is always imperfect. A “common language” effect size measure can describe the association of IQ phenotype with its genotype (Dunlap, 1994). [1] Suppose John scored higher on an IQ test than William. If heritability (h²) equals .50, then .75 is the probability that John’s genotype was more favorable to high IQ than William’s. If h² = .70, then the probability would rise to .81. John’s claim to a superior IQ genotype could be verified, of course, if each man had equally intelligent spouses and, say, 12 children – but few people would undertake this “test” willingly!

[1] The correlation of genotype and phenotype is the square root of heritability. A table in Dunlap’s (1994) article provides the translation of a correlation between X and Y into the probability that, if person A scores higher than person B on X, that person A also scores higher on Y.

Similar Findings From Early Intervention Studies

Both adoptions and early intervention programs can create new circumstances for children. Adoption may put a child into a social class that is very different from the situation of the biological parents. In early interventions, children are offered new kinds of environmental stimulation, although these interventions are created by program managers rather than by joining a new family.

Of the unintended intervention of adoption and the intended one of an in-school program, the latter is probably the weaker. Project Head Start, and similar programs, may give preschool children an earlier exposure to educational curriculum in the classroom. They sometimes also provide training to make parents more supportive of education and to encourage parents to help children progress educationally (e.g., by reading to their children). Head Start children, however, are usually destined to remain in poor neighborhoods for their elementary school education. They cannot escape from their own parents’ lack of education. In contrast, the biological child of a poor, working-class parent may be placed into an upper-middle-class home, with well-educated parents, excellent neighborhood schools, and peers who are ambitious and supportive of education. Given the strong treatment integrity of the adoptive study, it is all the more contrary to the conventional wisdom that its long-term effects on IQ variation (i.e., the lack of adoptive sibling correlations) are so weak.

As with the adoptive studies, the effects of early intervention on poor children’s IQs also “fade out” (Holden, 1990; Spitz, 1986). The “fade-out,” which has been extensively described in the literature on early intervention programs, is a boost in IQ immediately after a program and then an IQ decline by middle elementary school, with IQ often returning to the levels of untreated children. Although they are detected in different ways (mean changes vs. a reduction in the correlation for adoptive siblings), it is engaging that children in both early interventions and adoptive studies show a “fade-out” of environmental effects on IQ.

An understanding of both phenomena is needed to reconcile them. My interpretation is that intellectually enriching early experiences at a younger age develop a greater IQ without making the children genotypically brighter. Because “genotypes drive experience,” the earlier IQ gains would be erased later. (It should be borne in mind that absolute knowledge and skill levels increase for everyone throughout childhood and adolescence. IQ represents rank relative to other people. One can gain in knowledge without changing much, or at all, in rank in ability.)

In the case of early “interventions,” whether preschool programs or adoptions, more rapid growth in intellectual skills and knowledge does take place for the children. But it is not a growth rate that can be sustained, as happens with age, when intellectual demands become more complex. As they grow up, children’s accumulated knowledge and understanding reflects more closely their gene-based abilities and less the experiential residue of a year or two of intervention programming, or of an early childhood in the stimulating environment of an advantaged adoptive home.

This interpretation of program effects receives experimental support from another source, a research design that asked a slightly different question: “What is the effect of schooling ?” In a clever study, children were compared whose birthdays fell just before and just after the date for school entrance (Morrison, Smith, & Dow-Ehrensberger, 1995). A priori, we would not believe that one group of children was more “intelligent” than the other. Yet, those children who by virtue of their birth dates had received schooling outperformed their unschooled peers on a number of intellectual tests.

For example, Table 1 illustrates Morrison and his colleagues’ results for one outcome: reading achievement. Before entry into school, neither group of children differed much in reading ability (the apparent group difference of .90 SD was not statistically significant). Those children, however, who attended first grade rather than kindergarten bloomed in their reading achievement. The group difference now amounted to a huge gap of 2.6 SD. After both groups had finished another year of schooling, the pattern reversed again – the younger children, who were in Grade 1, nearly caught up in reading level with the older children, who were now in Grade 2, mean difference = .36 SD units, despite the latter group having twice as many years of reading instruction. Memory performance on several tasks also showed this pattern of large gains from first grade, followed by compensatory gains by the younger children as they themselves had first-grade schooling.

A Place at the Policy Table - Behavior Genetics and Estimates of Family Environmental Effects on IQ - Table 1

Schooling thus may power intellectual growth. To say that children would show the same intellectual skills and knowledge without schooling as with it would be ludicrous; however, despite the necessity of schooling, variation between children in intellectual growth can be highly heritable and little affected by long-term exposure to whatever family environment. In the adoption studies, advantaged families seem to provide the same kind of gain as schooling does for unschooled children, a jump start on intellectual growth, but without an ultimate improvement in the level finally attained.

The children in Head Start, and in other preschool programs, also may show equally impressive early IQ gains against their less-schooled peers. However, like the adoptive IQ gains, these seem to be temporary. In a sole intervention program in which IQ scores remained high (on a repeated use of the same IQ test), academic achievement failed to improve generally (Jensen, 1989). Rather than being contradictory, the adoptive family research design and preschool interventions may be interpreted in much the same way, as producing temporary effects on the rate of intellectual growth but not on its ultimate level. Both findings would discourage the idea that IQ is easily malleable through programs aimed specifically at low-performing children whose underlying capacity to learn may be unchanged by them.

Causes of Group Differences and Individual Differences: Two Realms or One?

The logic of Turkheimer’s argument against two realms of developmental processes, group differences and individual variation, can be summarized as follows. Developmental influences should apply only to individuals, one at a time, not to groups. Groups do not “receive” any developmental influence as a unit – group names are abstractions, not recipients of developmental processes. Hence, a group mean on a trait would be merely an average of the different developmental influences experienced by individuals within the category; there would not be “group-specific” developmental processes.

Consider any variable assumed to create minority versus majority group differences in IQ: Would it affect individuals differently within these groups? Think, for example, of racism directed toward individuals who are more African American in physical appearance. The range of darkness of skin color and “Africanness” of facial features is enormous within the Black population in the United States, all the more so because of the considerable mixture of genes of African and European origins in African Americans. If racists pick on individuals who appear more “African,” then this bias would create variation in exposure to discrimination among lighter and darker skinned African Americans. Also, some Black individuals may have little contact with Whites and thus less opportunity to encounter racism directly; some may be more sensitive to criticism than others; some may have, by chance, been victimized more hurtfully. Hence, the degree of “exposure to racism” should create individual variation in any trait it affects, in addition to affecting group means. Although African Americans were used in this example, the logic would apply to any physical feature that distinguished a minority group.

In a sophisticated cultural explanation of racial and ethnic differences, Ogbu (1987) recognized that the existence of variation implies that not all minority members experience the same degree of stigmatization or discrimination. Ogbu wrote: “Of course, not everyone feels this way. Some Black Americans do not identify with the oppositional identity and cultural frame of reference; some do so only marginally” (p. 165). Or, in another article, Ogbu (1994) attributed the academic success of some African Americans to their ability to disguise their academic work and perseverance by various strategies that deflect attention away from their achievements (e.g., by accepting the role of class clown).

In summary, minority-unique developmental processes that have been postulated (e.g., racism, minority-unique values, see Helms, 1992) should differ in psychological strength from one individual to another, either because one individual is more exposed than another, or because one individual resists the psychological influence more than another. As such, they should contribute both to individual variation and to group means.

A Test of the Two-Versus One-Realm Hypotheses: Searching for Factor X

If any minority-unique influence on psychological development exists, call it a Factor X, then it could affect psychological traits like IQ. When a causal system for one group contains a causal influence missing in another, it should have effects that permeate throughout a network of variables (Rowe, Vazsonyi, & Flannery, 19941995). Any minority-unique causal variable should increase the variance of each variable it touches. Suppose its variance was 50, whereas that of all other influences combined was 225. If the minority-unique variable were to affect IQ independently of these other known effects, then IQ variation in the minority group would be 275, not, as is typically observed, 225 (i.e., the square of 1 SD of an IQ test, 15). Minority-unique causes also would tend to change correlations. Any two variables “bridged” by a minority-unique variable, if a positive association, would correlate more highly in the minority group than in the majority.

My colleagues and I (Rowe, Vazsonyi, & Flannery, 1994) have searched extensively for minority-unique determinants of traits using covariance matrices. These included both outcome variables (e.g., IQ) and assumed influence variables (e.g., the stimulatory quality of the home environment); each matrix contained the associations among about 10 variables. The matrices came from data sources with a total of 3,392 African Americans, 1,766 Hispanics, and 8,582 Whites, and one other data source with 906 Asians. They were statistically alike; any differences among them could be explained as chance occurrences, not as a result of any systematic developmental processes. Indeed, in one analysis, we demonstrated that random subdivisions of the African American total sample were no more different from Whites than they were from one another.

Table 2 presents a 10 variable X 10 variable matrix from the Richmond Youth Project, a famous study of delinquent behavior (Hirschi, 1969). Correlations greater than .20 are shown in italics. Even without complex statistics, inspection of the rows of Table 2 shows a striking similarity of correlational pattern. The strongest correlations were between self-reports of parental behavior, for example, .56 in Whites and .71 in Blacks for supervision of father with supervision of mother. Note that for IQ, the correlations were also remarkably similar. In both races, IQ correlated with grades, Y = .35 in Whites and .35 in African Americans, and with official delinquency, r = -.22 in Whites and -.19 in African Americans. In both races, official and self-report delinquency bore a similar association (.31 and .28, respectively).

A Place at the Policy Table - Behavior Genetics and Estimates of Family Environmental Effects on IQ - Table 2

One way to interpret the amazing similarity of correlation (or covariance) matrices between populations is that only “one set of developmental processes” occurred within these populations in the United States. If this is so, then influences on individual variation should also be responsible for the mean differences. If the heritability of IQ were .40 or greater within groups, then it would seem reasonable to anticipate that differences in mean IQs for the different groups would have a partly genetic basis.

This line of argument, however, can be only suggestive, not definitive. MacKenzie (1984) has criticized it as the “hereditarian fallacy”:

First, it [the fallacy] assumes that our inability to identify the relevant [in our terms, minority-unique] variables provides good grounds for judging that there are none to be identified and therefore for rejecting the environmental hypothesis generally. (p. 1222)

Although MacKenzie’s reservation about the genetic hypothesis was properly taken, advances in research designs and statistical analyses now permit a more direct estimation of the extent to which the genetic and environmental sources of group means and individual variation are identical. The next section briefly describes one such approach.

The Group Differences Question: Modeling Both Means and Variation

The group differences question comes in two parts: (1) Do means and variation have the same determinants? (2) If so, what part of a mean difference is genetic or environmental? In the branch of statistics called “structural equation modeling,” recent advances permit models to consider both means and individual variation simultaneously (Joreskog & Sorbom, 1988).

We employed this approach in a study of African Americans’ and Whites’ academic achievement using as a sample children in the National Longitudinal Survey of Youth (NLSY). Three variables were taken to define an achievement factor: Peabody Individual Achievement Test (PIAT) reading recognition, PIAT reading comprehension, and PIAT mathematics (Rowe & Cleveland, 1996). Although our study did not have IQ subtests, the strong genetic correlation between IQ and achievement (Plomin & Petrill, 1997) would suggest results from one would generalize to the other.

The participants were full- and half-sibling pairs who were identified in the NLSY. Mothers’ reports of racial identity were used to distinguish African Americans and Whites. Covariance matrices were computed for the three tests. However, as three scores were available for two children in each family, they were 6 X 6. They contained a variety of associations: (a) the correlation of one test with another within persons, (b) the correlation of one test for siblings, and (c) the correlation of Test A for one sibling with Test B for the other.

The use of two levels of genetic relatedness permitted further the estimation of genetic and shared environmental factors. Figure 2 illustrates a part of the quantitative genetic model. Achievement was determined by both a genetic factor (G) with loadings a to c and by a shared environmental factor (SE) with loadings e to g. The model further required that the genetic factor correlate .50 in full siblings and .25 in half-siblings, whereas the family environmental factor was required to correlate 1.0 for both types of siblings.

A Place at the Policy Table - Behavior Genetics and Estimates of Family Environmental Effects on IQ - Figure 2

We first showed that individual variation had the same determinants for African Americans and Whites. This was done by determining whether factor loadings (see Figure 2) were equal across groups; they were.

In the second step, the factor loadings were fixed as constants in the quantitative genetic model. The group means were then added. If they had different determinants than did individual variation in achievement, it would be possible to reject the model statistically at this point. However, the model fit as well with the means added to it as it had without them.

The model-fitting procedure can estimate the contribution of genes and environment to racial group differences observed in the sample. Genes accounted for 66% to 74% of the observed group difference in verbal achievement and 36% of the difference in mathematics achievement. Shared environment accounted for the remainder, 34% to 26% of the difference in verbal achievement and 64% of that in mathematics achievement.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s