Everything is Correlated @lilithelina - Tumblr Blog

New Life, New Projects

Hello there!

I haven't posted here in two years (almost three looking at my last BCCIU post!), since it turned out that tumblr is not really my platform. I also was busy doing other things, like my PostDoc and producing another human...

I still hope to finish this project at some point, in some form, when I have a little more free time / me time, maybe. I will most likely do that at GitHub or GitLab, depending on whether or not I decide to leave GitHub, though, since I don't expect to be graded for coursera anymore now.

Until then, feel free to follow my current work project, for now still on GitHub - it deals with genome graph and is very interesting!

#BCCIU #genome graphs #news

•18+ Adults Only

Watch Anya Live on Cam

Anya is live and ready to show you everything. Watch her strip, dance, and perform exclusive shows just for you. Interact in real-time and make your fantasies come true.

✓ Live Streaming✓ Interactive Chat✓ Private Shows✓ HD Quality✓ Free Actions

Free to watch • No registration required • HD streaming

yourfluffiestnightmare

My friend told me a story he hadn’t told anyone for years. When he used to tell it years ago people would laugh and say, ‘Who’d believe that? How can that be true? That’s daft.’ So he didn’t tell it again for ages. But for some reason, last night, he knew it would be just the kind of story I would love. When he was a kid, he said, they didn’t use the word autism, they just said ‘shy’, or ‘isn’t very good at being around strangers or lots of people.’ But that’s what he was, and is, and he doesn’t mind telling anyone. It’s just a matter of fact with him, and sometimes it makes him sound a little and act different, but that’s okay. Anyway, when he was a kid it was the middle of the 1980s and they were still saying ‘shy’ or ‘withdrawn’ rather than ‘autistic’. He went to London with his mother to see a special screening of a new film he really loved. He must have won a competition or something, I think. Some of the details he can’t quite remember, but he thinks it must have been London they went to, and the film…! Well, the film is one of my all-time favourites, too. It’s a dark, mysterious fantasy movie. Every single frame is crammed with puppets and goblins. There are silly songs and a goblin king who wears clingy silver tights and who kidnaps a baby and this is what kickstarts the whole adventure. It was ‘Labyrinth’, of course, and the star was David Bowie, and he was there to meet the children who had come to see this special screening. ‘I met David Bowie once,’ was the thing that my friend said, that caught my attention. ‘You did? When was this?’ I was amazed, and surprised, too, at the casual way he brought this revelation out. Almost anyone else I know would have told the tale a million times already. He seemed surprised I would want to know, and he told me the whole thing, all out of order, and I eked the details out of him. He told the story as if it was he’d been on an adventure back then, and he wasn’t quite allowed to tell the story. Like there was a pact, or a magic spell surrounding it. As if something profound and peculiar would occur if he broke the confidence. It was thirty years ago and all us kids who’d loved Labyrinth then, and who still love it now, are all middle-aged. Saddest of all, the Goblin King is dead. Does the magic still exist? I asked him what happened on his adventure. ‘I was withdrawn, more withdrawn than the other kids. We all got a signed poster. Because I was so shy, they put me in a separate room, to one side, and so I got to meet him alone. He’d heard I was shy and it was his idea. He spent thirty minutes with me. ‘He gave me this mask. This one. Look. ‘He said: ‘This is an invisible mask, you see? ‘He took it off his own face and looked around like he was scared and uncomfortable all of a sudden. He passed me his invisible mask. ‘Put it on,’ he told me. ‘It’s magic.’ ‘And so I did. ‘Then he told me, ‘I always feel afraid, just the same as you. But I wear this mask every single day. And it doesn’t take the fear away, but it makes it feel a bit better. I feel brave enough then to face the whole world and all the people. And now you will, too. ‘I sat there in his magic mask, looking through the eyes at David Bowie and it was true, I did feel better. ‘Then I watched as he made another magic mask. He spun it out of thin air, out of nothing at all. He finished it and smiled and then he put it on. And he looked so relieved and pleased. He smiled at me. ‘'Now we’ve both got invisible masks. We can both see through them perfectly well and no one would know we’re even wearing them,’ he said. ‘So, I felt incredibly comfortable. It was the first time I felt safe in my whole life. ‘It was magic. He was a wizard. He was a goblin king, grinning at me. ‘I still keep the mask, of course. This is it, now. Look.’ I kept asking my friend questions, amazed by his story. I loved it and wanted all the details. How many other kids? Did they have puppets from the film there, as well? What was David Bowie wearing? I imagined him in his lilac suit from Live Aid. Or maybe he was dressed as the Goblin King in lacy ruffles and cobwebs and glitter. What was the last thing he said to you, when you had to say goodbye? ‘David Bowie said, ‘I’m always afraid as well. But this is how you can feel brave in the world.’ And then it was over. I’ve never forgotten it. And years later I cried when I heard he had passed.’ My friend was surprised I was delighted by this tale. ‘The normal reaction is: that’s just a stupid story. Fancy believing in an invisible mask.’ But I do. I really believe in it. And it’s the best story I’ve heard all year.

Paul Magrs (via yourfluffiestnightmare)

This is so beautiful.

90% of the work as a data scientist is cleaning the data

I've heard this - or similar statements - many times, and last weekend I fought with the perfect example, which I'd like to share.

Back in May (Oh God, it's already so long ago??) I started a fun project analysing exercise data I tracked with my Garmin fitness tracker. A few weeks in, I realised that, if I wanted to continue doing this every week until the middle of September, having a .csv file for each week to analyse could be... annoying. Incidentally, I also took a course in SQL database management, which inspired me to set up a little database for my exercise data.

It took me ages, being busy with other things, to even get started on that, but I've had the last two weeks off and finally got started. Last weekend, I wanted to try to upload a first batch of data. Said data was downloaded directly from the Garmin Connect website as .csv, over the format of which I have no control (even getting out a specific time range is difficult/impossible).

Of one problem I already knew from my previous analyses: the duration of running workouts is measured in minutes, seconds, and milliseconds, but not in MM:SS.mmm, but in MM:SS:mm. All other data is displayed either as MM:SS or HH:MM:SS, if applicable. I hope you see that this is awkward and terribly annoying. My previous solution was changing this manually each week, but I'm so far behind now, and it's uncomfortable anyway, that I switched to doing it in my Python script. The simple, first thought solution was to split the time string at the colons, add a "00:" at the beginning, and remove the millisecond value. This will of course stop working if I ever go for a run that's longer than an hour, but I don't plan to do that any time soon. I hope to find someone who does that with Garmin, though, to see what the format would look like then, and change my script accordingly.

The second problem I only discovered last weekend, and I'm honestly a bit shocked that anyone would do that. The table of course contains workout speed data, called "avg pace" and "max pace". Online, when viewing the table, the unit of this value is given in something like small print, for each workout. That's because Garmin is using very different values for different types of exercises. Running and walking are both provided in min/km, so I get values in MM:SS format here again. Cycling and uncategorised exercises, on the other hand, are listed with kph values, meaning they're simple decimals. Now, when I set up the database I forgot to check on this and decided that the "avg pace" and "max pace" columns should contain numeric values, and of course that blew up in my face. As far as I can see, I have three choices now:

Stick to this - what my gut calls - horrible concept of having values of different units in the same column, and just upload everything as strings for later parsing.

Create extra columns - to have two each for pace (in min/km) and speed (in kph), and only use the ones that work for that specific exercise. This means not only extra columns (two of which will always be NA while the other two are filled) but also extra work, making sure that everything is sorted correctly.

Convert either of the two formats into the other to have consistent data. This could then be converted back to the original format for analysis, if needed.

Oh, and there's also a fourth option, which is "Oh, f*** this, I didn't need speed/pace in my analysis so far, I'll just ignore that." I think I'll go with option three, though. ;-)

Conclusion

You would think that a widely used product that tracks workouts would provide you with suitable data to download, but I hope I made you see that this is not necessarily the case. Personally, I believe that saving data with different units (and formats!) in the same column goes against all common sense, but even here I was apparently wrong.

#SummerPain #data analysis #cleaning data #garmin

chrisriddellblog

Neil Gaiman on Libraries and Librarians.

lilithelina

This is beautiful.

micdotcom

Alison Chandra never expected to go viral for tweeting about her son’s medical bills — in fact, she never expected to be the mother of a child who needed so much medical care in the first place. But, with one viral tweetstorm she made it clear just how important the Affordable Care Act is to so many families. Read more (6/26/17)

lilithelina

US American politics are so weird. Why would you let people suffer and die, when it's so easy to help and support them? Why can nobody be expected to share their wealth? Why is "me and mine" more important than people in need?

•18+ Adults Only

Watch Anya Live on Cam

Anya is live and ready to show you everything. Watch her strip, dance, and perform exclusive shows just for you. Interact in real-time and make your fantasies come true.

✓ Live Streaming✓ Interactive Chat✓ Private Shows✓ HD Quality✓ Free Actions

Free to watch • No registration required • HD streaming

mcmansionhell

Going offline for a while

Zillow is threatening to sue me if I don’t delete most of the posts on this blog. Anyone who can see this who can help, please contact

[email protected]

It's already been three weeks since we started with #SummerPain! I'm still on track, I'm proud to say.

#SummerPain #exercise #rstats #ggplot #datavis

Multiple news outlets are running stories today about the role of alcohol in promoting breast cancer risk. ABC: Consuming just 1 alcoholic drink a day increases breast cancer risk, report finds Washington Post: Just one alcoholic drink a day increases risk of breast cancer, study says USA Today: One small drink a day increases breast …

It's a pity I don't have alcohol consumtion data as well in my project. ;-)

#correlation is not causation #breast cancer #alcohol #news

My week two data contains activities I tracked from May 21st to May 27th, again sorted into different types (cycling, running, etc.) together with relevant data, as well as some overall statistics for each day.

#SummerPain #datavis #exercise

I’m taking a break from my coursera project while we prepare to move to a bigger flat and I’m generally busy settling into my role as a postdoc with new projects on top of the old. Nevertheless, a little fun with programming at home can’t be wrong, right?

I'm taking a break from my coursera project while we prepare to move to a bigger flat and I'm generally busy settling into my role as a postdoc with new projects on top of the old. Nevertheless, a little fun with programming at home can't be wrong, right?

In summary - when counting cycling as exercise - I completed six workouts in the first week of #SummerPain. This brings me up to 6/100 workouts as of May 20th.

#SummerPain #ScientistsWhoYoga #datavis #exercise

•18+ Adults Only

Watch Anya Live on Cam

Anya is live and ready to show you everything. Watch her strip, dance, and perform exclusive shows just for you. Interact in real-time and make your fantasies come true.

✓ Live Streaming✓ Interactive Chat✓ Private Shows✓ HD Quality✓ Free Actions

Free to watch • No registration required • HD streaming

errantscience

Resist

iwillfightu

drained of blood, the heart is white

thepowerliftingpinay

woah

stan-leigh-bowery

No, that is NOT what this is. You’ve taken an amazing medical invention, a total game changer, and made up some stupid, faux-deep sentence fragment for it that is a complete falsehood. You should be embarrassed and ashamed, honestly.

This is a ghost heart. What they’ve done is taken a pig heart and stripped it down to, basically, a cell framework that they can use to BUILD A NEW HEART UPON. You could inject stem cells into this framework so that a newly formed personalized heart can be transplanted into a donor with a significantly reduced chance of rejection. FUCKING AMAZING. It’s not been done with human tissue yet, but the promise this given to people who need hearts - or kidneys or livers or whatever - is beautiful. Science is beautiful.

And it’s IMPERATIVE to mention that a woman, Doris Taylor, at the Texas Heart Institute developed this. And she started with a rat heart and worked up to he bigger, more complex (and more human) pig heart. What a total bad ass.

So look, quit making shit up, learn to do a reverse image search on stuff you find on the internet, and STOP ERASING WOMEN IN SCIENCE.

#science #ghost heart #transplantation #medicine

Logistic Regression - R

As mostly mentioned first, I want to compare Python and R analysis steps in the DataManViz, DataAnaT, and RegModPrac projects. Therefore, this is the R version of the Logistic Regression Python script I posted before. Here, I'll use logistic regression to test the association between internet use rate (my response variable, this time binned into two categories) and multiple explanatory variables - but first and foremost new breast cancer cases. Again, the whole thing will look better over here. I had to switch back to using RMarkdown, since Jupyter had some problems I do not yet understand.

I will first run some of my previous code to remove variables I don't need and observations for which important data is missing.

# load data gapminder <- read.table("../gapminder.csv", sep = ",", header = TRUE, quote = "\"") # set row names rownames(gapminder) <- gapminder$country # subset data sub_data <- subset(gapminder, select = c("breastcancerper100th", "urbanrate", "internetuserate", "incomeperperson")) # remove rows with NAs sub_data2 <- na.omit(sub_data)

Internet usage, my response variable, has to be in a binary format for logistic regression to work. In my Python script, I decided to use the 25% quartile (9.1) as cut-off, so I'll do the same here.

# bin response variable sub_data2$internetBin <- as.numeric(sub_data2$internetuserate > 9.1) summary(sub_data2)

breastcancerper100th urbanrate internetuserate incomeperperson Min. : 3.90 Min. : 10.40 Min. : 0.720 Min. : 103.8 1st Qu.: 20.60 1st Qu.: 36.84 1st Qu.: 9.102 1st Qu.: 691.1 Median : 30.30 Median : 59.46 Median :28.732 Median : 2425.5 Mean : 37.78 Mean : 56.25 Mean :33.747 Mean : 7312.4 3rd Qu.: 50.35 3rd Qu.: 73.49 3rd Qu.:52.513 3rd Qu.: 8880.4 Max. :101.10 Max. :100.00 Max. :95.638 Max. :52301.6 internetBin Min. :0.0000 1st Qu.:0.5000 Median :1.0000 Mean :0.7485 3rd Qu.:1.0000 Max. :1.0000

For the binning, I'm using a simple test whether the internet usage in a country is above my threshold or not. The as.numeric() will automatically convert every returned TRUE to a 1, and every FALSE to a 0 -- and done is my conversion.

Let's go ahead with the logistic regression!

fit1 <- glm(internetBin ~ breastcancerper100th, data = sub_data2, family = "binomial") summary(fit1)

Call: glm(formula = internetBin ~ breastcancerper100th, family = "binomial", data = sub_data2) Deviance Residuals: Min 1Q Median 3Q Max -1.9427 -0.3112 0.2517 0.7259 1.7188 Coefficients: Estimate Std. Error z value Pr(>|z|) (Intercept) -1.89496 0.57693 -3.285 0.00102 ** breastcancerper100th 0.10578 0.02285 4.630 3.66e-06 *** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 (Dispersion parameter for binomial family taken to be 1) Null deviance: 183.87 on 162 degrees of freedom Residual deviance: 135.78 on 161 degrees of freedom AIC: 139.78 Number of Fisher Scoring iterations: 6

The model results look quite different from those returned by Python. Already familiar are the coefficients shown in the middle of the results, which say that there is a significant, positive association between internet usage and breast cancer. From the parameter estimate, the odds ratio can again be calculated, but for that it might be nice to have the confidence intervals as well. Luckily, these are relatively easy to obtain:

conf1 <- confint(fit1)

Waiting for profiling to be done...

conf1 <- cbind(conf1, OR = coef(fit1)) print("odds ratio with confidence intervals") exp(conf1)

[1] "odds ratio with confidence intervals" 2.5 % 97.5 % OR (Intercept) 0.04444601 0.432423 0.1503249 breastcancerper100th 1.06804211 1.168457 1.1115793

The values are the same as in Python, so the OR > 1 again signifies that internet usage will be higher in countries with higher breast cancer prevalence.

Below the coefficients, the R function lists deviance values. I am not really sure what they signify in the context of logistic regression, but apparently the goal of the model is to reduce them as much as possible. How successful the model is at that task can be analysed with an a deviance table using the Chi square test to calculate test statistics:

anova(fit1, test = "Chisq")

Analysis of Deviance Table Model: binomial, link: logit Response: internetBin Terms added sequentially (first to last) Df Deviance Resid. Df Resid. Dev Pr(>Chi) NULL 162 183.87 breastcancerper100th 1 48.088 161 135.78 4.076e-12 *** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

The null model uses only the intercept and no explanatory variable, resulting in a high deviance. With breast cancer as explanatory variable, a significant reduction of deviance takes place. If the model would contain more explanatory variables, the test would add them one at a time and determine their effect on the deviance. Time to add another explanatory variable!

fit2 <- glm(internetBin ~ breastcancerper100th + incomeperperson, data = sub_data2, family = "binomial") summary(fit2)

Call: glm(formula = internetBin ~ breastcancerper100th + incomeperperson, family = "binomial", data = sub_data2) Deviance Residuals: Min 1Q Median 3Q Max -2.50062 -0.31160 0.08903 0.58516 1.78481 Coefficients: Estimate Std. Error z value Pr(>|z|) (Intercept) -2.0100509 0.6218483 -3.232 0.00123 ** breastcancerper100th 0.0761711 0.0256152 2.974 0.00294 ** incomeperperson 0.0004431 0.0001703 2.602 0.00927 ** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 (Dispersion parameter for binomial family taken to be 1) Null deviance: 183.87 on 162 degrees of freedom Residual deviance: 120.18 on 160 degrees of freedom AIC: 126.18 Number of Fisher Scoring iterations: 8

In Python, using both breast cancer and income as explanatory variables resulted in a warning about quasi complete separation of the two internet usage groups. This is not the case here, I wonder why...

Let's have a look at the odds ratios and deviance again...

conf2 <- confint(fit2)

Waiting for profiling to be done...

Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred Warning: glm.fit: fitted probabilities numerically 0 or 1 occurred

conf2 <- cbind(conf2, OR = coef(fit2)) print("odds ratio with confidence intervals") exp(conf2)

[1] "odds ratio with confidence intervals" 2.5 % 97.5 % OR (Intercept) 0.03603527 0.4218608 0.1339819 breastcancerper100th 1.03010585 1.1398341 1.0791472 incomeperperson 1.00016777 1.0008310 1.0004432

Ah, here we go! This warning indicates that there is (quasi) complete separation happening, so I'll just move on to my third model, using breast cancer and urbanisation as explanatory variables.

fit3 <- glm(internetBin ~ breastcancerper100th + urbanrate, data = sub_data2, family = "binomial") summary(fit3)

Call: glm(formula = internetBin ~ breastcancerper100th + urbanrate, family = "binomial", data = sub_data2) Deviance Residuals: Min 1Q Median 3Q Max -2.1845 -0.2828 0.2272 0.5919 1.9885 Coefficients: Estimate Std. Error z value Pr(>|z|) (Intercept) -3.38831 0.72676 -4.662 3.13e-06 *** breastcancerper100th 0.07813 0.02318 3.370 0.000752 *** urbanrate 0.04759 0.01250 3.808 0.000140 *** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 (Dispersion parameter for binomial family taken to be 1) Null deviance: 183.87 on 162 degrees of freedom Residual deviance: 118.51 on 160 degrees of freedom AIC: 124.51 Number of Fisher Scoring iterations: 6

All right, what do the additional tests say this time?

conf3 <- confint(fit3)

Waiting for profiling to be done...

conf3 <- cbind(conf3, OR = coef(fit3)) print("odds ratio with confidence intervals") exp(conf3)

[1] "odds ratio with confidence intervals" 2.5 % 97.5 % OR (Intercept) 0.007270543 0.1280973 0.03376562 breastcancerper100th 1.037775543 1.1373111 1.08126415 urbanrate 1.024544428 1.0763632 1.04873683

print("Chi-squared test of deviance")

[1] "Chi-squared test of deviance"

anova(fit3, test = "Chisq")

Analysis of Deviance Table Model: binomial, link: logit Response: internetBin Terms added sequentially (first to last) Df Deviance Resid. Df Resid. Dev Pr(>Chi) NULL 162 183.87 breastcancerper100th 1 48.088 161 135.78 4.076e-12 *** urbanrate 1 17.267 160 118.51 3.247e-05 *** --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

The odds ratio for an association between breast cancer and internet usage decreased a bit when urbanisation was added to the model -- as before. On the other hand, the residual deviance also decreases significantly upon addition of that second explanatory variable, so it definitely improves the model, while not being a confounder (since breast cancer is still significantly associated with internet usage).

We can also subtract the null deviance from the other residual deviances calculated here, to have a simple view of the effect of the different explanatory variables. A high deviance is problematic, and since usually the null deviance (with the model using only the intercept) is higher than the deviance when explanatory variables are included, the differences shows just how much the model improved upon the addition of the variable.

print("difference in residual deviance using only breast cancer") with(fit1, null.deviance - deviance) print("difference in residual deviance using breast cancer and internet usage") with(fit2, null.deviance - deviance) print("difference in residual deviance using breast cancer and urbanisation") with(fit3, null.deviance - deviance)

[1] "difference in residual deviance using only breast cancer" [1] 48.08763 [1] "difference in residual deviance using breast cancer and internet usage" [1] 63.68923 [1] "difference in residual deviance using breast cancer and urbanisation" [1] 65.35506

The addition of other explanatory variables definitely improved my logistic model, but overall I think it's nowhere near the multiple linear regression which can directly work with my continuous data.

#RegModPrac #BCCIU #Logistic Regression #R

Why do these things correlate? These 15 correlations will blow your mind. (Is this headline sensationalist enough for you to click on it yet?)

Everything is correlated. ;-)

#correlation #correlation is not causation #spurious correlation #fun with statistics

The first time it happened he had analyzed a dataset and the results were just shy of significance. Fox noticed that if he duplicated a couple of cases and deleted a couple of cases, he could shift the p-value to below .05. And so he did. Fox recognized that the system rewarded him, and his collaborators, not for interesting research questions, or sound methodology, but for significant results.

"p < 0.05" is a problem for science. It's a bit like an addiction - you get rewarded for having it, and not even looking for it is out of the question in many cases. It will take years and a lot of adjustments to get rid of something as ingrained into the scientific process as the p-value, but I hope that, the more we emphasise the problems, the easier it will become to refuse to use or request it.

#p-value #significance #science #fraud

•18+ Adults Only

Watch Anya Live on Cam

Anya is live and ready to show you everything. Watch her strip, dance, and perform exclusive shows just for you. Interact in real-time and make your fantasies come true.

✓ Live Streaming✓ Interactive Chat✓ Private Shows✓ HD Quality✓ Free Actions

Free to watch • No registration required • HD streaming

explore-blog

I am impatient with people like Richard Dawkins who try to disprove the existence of God with science. Science can never prove or disprove the existence of God. If God exists, God surely exists outside of rational thought and outside the physical universe.

Speaking about science and religion at the Jewish Theological Seminary on October 16, 2013, Alan Lightman – one of the greatest science writers and storytellers of our time, and the very first person to receive dual appointments in science and the humanities at MIT – offers a “working definition of God” as “a being that lives outside the lines of time and space and is not restricted by the laws of the physical universe,” a definition that automatically negates any effort to prove or disprove the existence of God via physical laws.

Lightman added that “science and religion share a sense of wonder.”

Complement with Carl Sagan on science and spirituality, Isaac Asimov on religion vs. humanism, and Albert Einstein on whether scientists pray.

(via patsywalker)

#SO WEIRD HOW THOSE ARE ALL JEWS

what a ~wacky coincidence~ lmao

p>(via phyteclub)

#great quote #richard dawkins #alan lightman #religion and science

microculture

We now know that antibacterial soaps do not protect from preventable illness better than regular soap, but beyond that, some of these products are even shown to have deleterious long-term effects.

lilithelina

Please just use normal soap - it's (probably) better for your wallet and it is better for you.

#antibacterial soap #antibiotics #false advertising #antibiotic resistance

Trending Blogs

Last Seen Blogs

Everything is Correlated